Datasets:

kejian

/

SciReviewGen

like 0

Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> A. ULTRA-RELIABILITY <s> Gupta and Kumar (2000) introduced a random network model for studying the way throughput scales in a wireless network when the nodes are fixed, and showed that the throughput per source-destination pair is /spl otimes/(1//spl radic/nlogn). Grossglauser and Tse (2001) showed that when nodes are mobile it is possible to have a constant or /spl otimes/(1) throughput scaling per source-destination pair. The focus of this paper is on characterizing the delay and determining the throughput-delay trade-off in such fixed and mobile ad hoc networks. For the Gupta-Kumar fixed network model, we show that the optimal throughput-delay trade-off is given by D(n) = /spl otimes/(nT(n)), where T(n) and D(n) are the throughput and delay respectively. For the Grossglauser-Tse mobile network model, we show that the delay scales as /spl otimes/(n/sup 1/2//v(n)), where v(n) is the velocity of the mobile nodes. We then describe a scheme that achieves the optimal order of delay for any given throughput. The scheme varies (i) the number of hops, (ii) the transmission range and (iii) the degree of node mobility to achieve the optimal throughput-delay trade-off. The scheme produces a range of models that capture the Gupta-Kumar model at one extreme and the Grossglauser-Tse model at the other. In the course of our work, we recover previous results of Gupta and Kumar, and Grossglauser and Tse using simpler techniques, which might be of a separate interest. <s> BIB001 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> A. ULTRA-RELIABILITY <s> Wireless 5G systems will not only be "4G, but faster". One of the novel features discussed in relation to 5G is Ultra-Reliable Communication (URC), an operation mode not present in today's wireless systems. URC refers to provision of certain level of communication service almost 100 % of the time. Example URC applications include reliable cloud connectivity, critical connections for industrial automation and reliable wireless coordination among vehicles. This paper puts forward a systematic view on URC in 5G wireless systems. It starts by analyzing the fundamental mechanisms that constitute a wireless connection and concludes that one of the key steps towards enabling URC is revision of the methods for encoding control information (metadata) and data. It introduces the key concept of Reliable Service Composition, where a service is designed to adapt its requirements to the level of reliability that can be attained. The problem of URC is analyzed across two different dimensions. The first dimension is the type of URC problem that is defined based on the time frame used to measure the reliability of the packet transmission. Two types of URC problems are identified: long-term URC (URC-L) and short-term URC (URC-S). The second dimension is represented by the type of reliability impairment that can affect the communication reliability in a given scenario. The main objective of this paper is to create the context for defining and solving the new engineering problems posed by URC in 5G. <s> BIB002 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> A. ULTRA-RELIABILITY <s> Due to its flexibility, cost-efficiency, and the ability to support mobility, wireless connectivity is seen today as a key enabler for a wide range of applications beyond classical mobile communications. A significant part of these applications depends on the capability of the wireless communication system to provide reliable connectivity. However, due to the randomness of the wireless propagation channel, reliability is still a critical issue in these systems. Some applications, such as vehicular and industrial applications, demand a level of reliability that wireless communication systems typically are not able to guarantee. This paper provides a framework that enables these applications to make use of wireless connectivity only if the transmission conditions are favorable enough. The concept is based on the idea that - despite the fact that it is practically impossible to ensure error-free wireless communication - it is feasible to derive boundary conditions for the transmission success. To this end, the paper introduces a novel metric for Ultra-Reliable Communication (URC) referred to as “Availability”, that determines the expected presence or absence of link reliability at the time of transmission. The availability is signaled by means of an Availability Indicator (AI) to the applications. Moreover, we develop the system model for computing the AI and illustrate the potential benefits of the new reliability metric by means of a possible implementation for automotive scenarios. <s> BIB003 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> A. ULTRA-RELIABILITY <s> The fifth generation (5G) of cellular networks is starting to be defined to meet the wireless connectivity demands for 2020 and beyond. One area that is considered increasingly important is the capability to provide ultra-reliable and low-latency communication, to enable e.g., new mission-critical machine-type communication use cases. One such example with extremely demanding requirements is the industrial automation with a need for ultra-low latency with a high degree of determinism. In this paper, we discuss the feasibility, requirements and design challenges of an OFDM based 5G radio interface that is suitable for mission-critical MTC. The discussion is further accompanied with system-level performance evaluations that are carried out for a factory hall-wide automation scenario with two different floor layouts. <s> BIB004 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> A. ULTRA-RELIABILITY <s> 5G is envisioned to support unprecedented diverse applications and services with extremely heterogeneous performance requirements, i.e., mission critical IoT communication, massive machine-type communication and Gigabit mobile connectivity. This imposes enormous challenges to fulfil the key performance requirements, in particular, mission critical IoT communication, which calls for a dramatic paradigm change in 5G. This paper presents vision and challenges of mission critical IoT scenarios and the enabling technologies in 5G. Several research opportunities are given as example for inspiration purpose. <s> BIB005 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> A. ULTRA-RELIABILITY <s> 5G is envisioned to support three broad categories of services: eMBB, URLLC, and mMTC. URLLC services refer to future applications which require reliable data communications from one end to another, while fulfilling ultra-low latency constraints. In this paper, we highlight the requirements and mechanisms that are necessary for URLLC in LTE. Design challenges faced when reducing the latency in LTE are shown. The performance of short processing time and frame structure enhancements are analyzed. Our proposed DCI Duplication method to increase LTE control channel reliability is presented and evaluated. The feasibility of achieving low latency and high reliability for the IMT-2020 submission of LTE is shown. We further anticipate the opportunities and technical design challenges when evolving 3GPP's LTE and designing the new 5G NR standard to meet the requirements of novel URLLC services. <s> BIB006

Reliability denotes to the capability of guaranteeing successful message transmissions from one peer to another within a given latency bound BIB004 , BIB002 . The work presented in BIB003 , the reliability is separated into two parts: link reliability and system reliability. They have defined the link reliability as the data that can be transmitted successfully within a given time frame, while system reliability is defined as when a system can precisely indicate the absence of link reliability and ensure the presence of link reliability when required by the application. Reliable communication has been well-studied as a fundamental problem in the information theory. Shannon's landmark theoretical study shows that the possibility of communication with vanishing probability of error is at non-zero rates BIB001 . Reliability has been studied as well from the networking perspective to complement the techniques used at PHY layer. The reliability problem is a challenge in MTC context without human assistance, due to long distance transmissions, intermediate routing, and wireless tampering and sniffing. However, the MTC requirements differ for various types of services BIB004 . The most recent 4G systems can achieve a reliability of approximately 10 −2 and measure as Typical Block Error Rate (BLER) BIB005 , while the performance requirements of mcMTC services should be much better than this, and in fact, mcMTC in 5G services require reliability at least 10 −5 BLER within 1 ms of latency period , BIB006 . In respect of mcMTC services, the target BLER vary from 10 −5 to 10 −9 . However, according to the work conducted in BIB005 , the targeted reliability requirement of 5G will be 10 −9 residual BLER within the latency bound. Reliability measurement (BLER) can be formulated as: BLER is a function of Bit Error Rate (BER) given as: where N is the number of bits in the block. From equation 2, it can be clearly observed that the bigger block will lead to higher BLER. Therefore, it is recommended to use short code block to support providing ultra-reliable mcMTC services. 5G tends to provide more dynamic solutions to increase the reliability instead of providing solutions to increase the bandwidth as in LTE. For instance, a selective blanking of the strongest interferers during retransmissions could significantly boost the success probability. Considering a different level of reliability for different mcMTC services, we need a way, in which a specific communication service is composed. As suggested by BIB003 , it is not necessary to fail the service whenever the reliability requirement is not fulfilled. Therefore, Reliable Service Composition (RSC) should offer a specific level of functionality when it is not possible to meet the full functionality. The key issue of making RSC operational in the case of MTC is by providing a reliable criteria for detecting the service level that the system should apply at a given time. The design of data/metadata for every service level should be integrated into an overall protocol, which could flexibly switch between modes as dictated by the dynamic conditions BIB002 . In summary, the implementation of RSC relies on a careful consideration of the requirements set by the application and the availability indicator that the communication layer provides to the application layer. Most relevant mcMTC reliability enablers are listed below and are discussed further in section V: 1) Waveform selection 2) Multiplexing scheme 3) Channel estimation accuracy improvement. 4) Redesign the resource allocation at MAC layer 5) Channel coding schemes. 6) Frame structures and short TTI 7) Network architecture including network slicing, caching, CloudRAN, and edge computing. 8) Antenna, space, and frequency diversity schemes 9) Supporting the prompt transmission of mcMTC packets by following multiplexing approach giving the priority to mcMTC packets over eMBB and mMTC packets. 10) Using Hybrid-ARQ (HARQ) on the MAC layer and erasure coding coupled with ARQ on the upper layers.

Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> B. ULTRA-LOW LATENCY <s> Communication technologies of the Tactile Internet have to achieve a combination of extremely low latency under high reliability and security constraints. The targeted applications are in the fields of industry automation and transport systems, healthcare, education and gaming. 5G addresses tactile use cases under the term mission-critical machine type communication. We demonstrate a first implementation of a wireless broadband communication system utilizing 20MHz bandwidth which can achieve a round-trip delay below 1 ms. Delay measurements were performed between two endpoints over the air. The system is based on a flexible Software Defined Radio (SDR) toolkit with PHY and MAC signal processing algorithms implemented on a multi-core DSP platform. The demonstration shows a live end-to-end transmission of real data packets while at the same time evaluating latency probes. In addition, we show the general feasibility of real-time implementation of ultra-low delay signal processing on SDRs. Finally, we give results by evaluation of latency and reliability performance indicators. <s> BIB001 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> B. ULTRA-LOW LATENCY <s> Next generation mobile networks not only envision enhancing the traditional MBB use case but also aim to meet the requirements of new use cases, such as the IoT. This article focuses on latency critical IoT applications and analyzes their requirements. We discuss the design challenges and propose solutions for the radio interface and network architecture to fulfill these requirements, which mainly benefit from flexibility and service-centric approaches. The article also discusses new business opportunities through IoT connectivity enabled by future networks. <s> BIB002 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> B. ULTRA-LOW LATENCY <s> The fifth generation (5G) wireless network technology is to be standardized by 2020, where main goals are to improve capacity, reliability, and energy efficiency, while reducing latency and massively increasing connection density. An integral part of 5G is the capability to transmit touch perception type real-time communication empowered by applicable robotics and haptics equipment at the network edge. In this regard, we need drastic changes in network architecture including core and radio access network (RAN) for achieving end-to-end latency on the order of 1 ms. In this paper, we present a detailed survey on the emerging technologies to achieve low latency communications considering three different solution domains: 1) RAN; 2) core network; and 3) caching. We also present a general overview of major 5G cellular network elements such as software defined network, network function virtualization, caching, and mobile edge computing capable of meeting latency and other 5G requirements. <s> BIB003 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> B. ULTRA-LOW LATENCY <s> URLLC is a new service category in 5G to accommodate emerging services and applications having stringent latency and reliability requirements. In order to support URLLC, there should be both evolutionary and revolutionary changes in the air interface named 5G NR. In this article, we provide an up-to-date overview of URLLC with an emphasis on the physical layer challenges and solutions in 5G NR downlink. We highlight key requirements of URLLC and then elaborate the physical layer issues and enabling technologies including packet and frame structure, scheduling schemes, and reliability improvement techniques, which have been discussed in the 3GPP Release 15 standardization. <s> BIB004

Latency refers to the time delay between the time at which data is generated and transmitted from one device, and the time at which the same data is correctly received by another device. According to 3GPP, latency in the network can be classified into control plane latency and user plane latency . • User plane latency also known as transport delay is defined as the one-way time it takes to successfully deliver an application layer packet/message from the radio protocol layer ingress point to the radio protocol ingress point of the radio interface, in either Uplink (UL) or Downlink (DL) in the network for a given service in unloaded conditions, assuming the mobile station is in the active state. • Control plane latency is defined as the transition time from a most ''battery efficient'' state (e.g., idle state) to the start of continuous data transfer (e.g. active state). The minimum requirements for user plane latency are 4 ms for eMBB and 1 ms for mcMTC assuming a single user while the minimum requirement for control plane latency is 20 ms . Proponents are encouraged to consider lower control plane latency, e.g. 10 ms . This requirement is defined for the purpose of evaluation in the eMBB and mcMTC usage scenarios. 5G community generally considers the network latency as one of the key requirements for future wireless networks enabling new applications by means of E2E latency that goes down to the millisecond level. Such latency cannot be performed by the current 4G technology. In this regard, we need drastic changes in network architecture including core and Radio Access Network (RAN) for achieving E2E latency on the order of 1 ms BIB003 . The current latency in LTE-A is about 5ms , BIB002 . Therefore, it is expected that the latency target for a new 5G radio access technology is going to be up to five times lower than compared with the latency of 4G networks (i.e. less or equal to 1ms). Researchers are wondering if the 1ms latency target will actually become a part of the 5G standard due to the restriction of fundamental laws of physics, such as the speed of light, the intrinsic properties of the fiber optics in which the light propagation within fiber optics is slower compared with the light propagation within vacuum. E2E latency T L is the time which includes the following components: processing time on the transmitter/receiver devices T proc , over-the-air transmission delay in one way T trans , and the network processing time T netproc (I.e. the base station (BS) and control server) . Taking 1ms E2E latency T L for instance, the distribution of the latency over various components can be as follows: 300µs for T proc , 200µs for T trans in both sides, and 500µs for T netproc . The processing latency, including the processing in the device T proc and the processing in the network T netproc , includes time for receiving data symbols, acquiring channel information, extracting control (scheduling) information, decoding data packet, and checking the existence of error BIB004 . Another elementary delay component is the Transmit Time Interval (TTI), defined as the minimum data block length, which is involved in each transmission of grant, data, and retransmission due to errors detected in higher layer protocols. Figure 5 assumes that every 100 bit packet must be transmitted within a 1ms E2E latency budget, as aimed to 100µs for over-the-air transmission time that corresponds to 10% of the E2E latency budget . However, the work presented in suggests that T L ≤ 1 ms to support mcMTC. FIGURE 5. Distribution of the E2E latency assuming a 100µs air interface delay . mcMTC services with ultra-low latency mostly transport very short control and command messages with the size of KBytes and Bytes and these messages are to be delivered very frequently. However, there are a few key reasons causing delay in RAN which are basically related to the following: 1) The packet retransmissions caused by channel errors and congestion. 2) The Link establishment caused by grant acquisition and random access . 3) Packet processing and queuing delay. 4) Channel coding delay. Therefore, it requires careful redesign of PHY and MAC layers to overcome these reasons. A clean slate redesign includes: • Requirement latency can be addressed based on 3GPP, by removing the retransmission by having early channel estimation and on-the-fly decoding . • Network architecture including CloudRAN and new entities such as Software Defined Networking (SDN), Network Functions Virtualization (NFV), Mobile Edge Cloud (MEC), and fog network along with new backhaul based solutions. • Caching solutions including caching placement, content delivery, centralized caching, and distributed caching BIB003 , BIB001 . • Minimizing system overhead, as an example, procedures for user scheduling, channel training, and resource allocation have to be combined or removed BIB004 . • Since the packet re-transmission mechanism degrades the latency, the probability of packet error of the first transmission has to be reduced significantly. • To avoid queuing delays at the radio transmitter the MAC layer should be designed to enable immediate access. This can be achieved by providing instant-access resource allocations dimensioned (i.e. real-time resource allocation) to minimize collision risks. • Maintaining multiple connectivity links simultaneously can provide diversity and redundancy to address such stringent requirements . • Reducing TTIs and widening the bandwidth of radio resource blocks to lower the latency over the radio link. This can be achieved by using fewer OFDM symbols per TTI and shortening OFDM symbols via wider Subcarrier Spacing (SCS) as well as lowering HARQ RTT . • mcMTC packet should be transmitted immediately without delay by following a multiplexing mechanism giving the priority to mcMTC packets. • For some use cases, low-latency communications are required between devices in close proximity. In this case, a direct D2D communication link can help in providing low-latency transmission . • Reducing queuing time by using fewer OFDM symbol duration causing an increase of the SCS and hence fewer resource blocks are available in the frequency domain. This shortcoming can be alleviated using grant-free transmission in the UL. On the DL, longer TTI is needed at high offered loads to cope with non-negligible queuing delays . • Provide a new/modified frame or packet structure, waveform designs, and multiple access techniques. • Modulation and coding schemes to reduce processing delays. • Control channels enhancements. • Other enabling techniques including modifying the spacing of the sub-carrier, providing low latency symbol detection, and mmWave aggregation. Recently, various low-latency transmission protocols have been suggested, such as grant-free, one-shot, and feedback-less protocols BIB004 . These are proposed to reduce the latency caused by the control signaling. The work conducted in BIB001 demonstrates the first wireless broadband communication system which can achieve an E2E latency below 1ms and 20MHz bandwidth between two endpoints over the air. The system is introduced based on the PHY and MAC signal processing algorithms, which are implemented on a multi-core demand-side platform with a flexible software-defined radio toolkit. The demonstration shows a transmission of real data packets while evaluating E2E latency probes at the same time. However, the demonstration does not mention how to cope up achieved latency with ultra-reliability needed for mcMTC systems. In BIB003 , a comprehensive survey is provided for low-latency solutions in the context of 5G from three different perspectives: Core network solutions, RAN solutions, and Caching solutions. However, the study does not provide a detailed comparison of these solutions. Lower latency concern took up its attention of 3GPP Release 15 by introducing two enhancements over LTE-A. The first is related to providing a shorter TTI. Traditional LTE TTI is 14 symbols which is 1ms scheduling interval. However, with TTI, both 7 symbols (0.5 ms) as well as 2 symbols (0.142 ms), scheduling interval is supported. The second is by reducing processing time by making the terminal respond to DL data and UL grants in 3ms instead of 4ms.

Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> C. AVAILABILITY <s> Due to its flexibility, cost-efficiency, and the ability to support mobility, wireless connectivity is seen today as a key enabler for a wide range of applications beyond classical mobile communications. A significant part of these applications depends on the capability of the wireless communication system to provide reliable connectivity. However, due to the randomness of the wireless propagation channel, reliability is still a critical issue in these systems. Some applications, such as vehicular and industrial applications, demand a level of reliability that wireless communication systems typically are not able to guarantee. This paper provides a framework that enables these applications to make use of wireless connectivity only if the transmission conditions are favorable enough. The concept is based on the idea that - despite the fact that it is practically impossible to ensure error-free wireless communication - it is feasible to derive boundary conditions for the transmission success. To this end, the paper introduces a novel metric for Ultra-Reliable Communication (URC) referred to as “Availability”, that determines the expected presence or absence of link reliability at the time of transmission. The availability is signaled by means of an Availability Indicator (AI) to the applications. Moreover, we develop the system model for computing the AI and illustrate the potential benefits of the new reliability metric by means of a possible implementation for automotive scenarios. <s> BIB001 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> C. AVAILABILITY <s> mcMTC is starting to play a central role in the industrial Internet of Things ecosystem and have the potential to create high-revenue businesses, including intelligent transportation systems, energy/ smart grid control, public safety services, and high-end wearable applications. Consequently, in the 5G of wireless networks, mcMTC have imposed a wide range of requirements on the enabling technology, such as low power, high reliability, and low latency connectivity. Recognizing these challenges, the recent and ongoing releases of LTE systems incorporate support for lowcost and enhanced coverage, reduced latency, and high reliability for devices at varying levels of mobility. In this article, we examine the effects of heterogeneous user and device mobility -- produced by a mixture of various mobility patterns -- on the performance of mcMTC across three representative scenarios within a multi-connectivity 5G network. We establish that the availability of alternative connectivity options, such as D2D links and drone-assisted access, helps meet the requirements of mcMTC applications in a wide range of scenarios, including industrial automation, vehicular connectivity, and urban communications. In particular, we confirm improvements of up to 40 percent in link availability and reliability with the use of proximate connections on top of the cellular-only baseline. <s> BIB002

Availability refers to the system endurance against possible outage scenarios. An mcMTC service should get a prompt response from the wireless medium to transmit the mcMTC packets immediately within a given time frame in the scheduling period (i.e., reserved resources for the mMTC and eMBB services have to be used for the mcMTC service). Based on the definition of reliability provided in BIB001 , reliability requires a Reliable Transmission Link (RTL). An Availability Estimation and Indication (AEI) mechanism has been proposed in BIB001 . This mechanism could predict the availability of RTL under certain conditions. The procedure is summarized as follows: the application sends an availability request to the AEI; then, the AEI evaluates the Signal-toInterference and Noise Ratio (SINR) and/or the Acknowledgment (ACKN)/Negative-ACKN (NACK) statistics of the retransmission protocols used at the link level. After that, the AEI provides an availability indicator for the application based on the original requirements of the availability request. However, this is at the cost of reducing the availability of RTL. Using retransmission improves the RTL but at the cost of increasing the latency. The authors in BIB002 explain the need to offer an availability about 10 −9 toward the next generation mcMTC systems, to control the probabilities of the underlying rare events. However, enhancing coverage is an essential issue for the deployment of a wireless mcMTC, in which devices and machines are installed in difficult locations. There are many solutions to enhance coverage, such as leveraging protection and redundancy capabilities which already have been deployed in most service provider networks, such as G.8032 Ethernet rings. In addition, enhanced availability can be gained by using repetition or sub-frame bundling, retransmission using HARQ and many others.

Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> V. MCMTC ENABLING TECHNOLOGIES <s> 5G is envisioned to support unprecedented diverse applications and services with extremely heterogeneous performance requirements, i.e., mission critical IoT communication, massive machine-type communication and Gigabit mobile connectivity. This imposes enormous challenges to fulfil the key performance requirements, in particular, mission critical IoT communication, which calls for a dramatic paradigm change in 5G. This paper presents vision and challenges of mission critical IoT scenarios and the enabling technologies in 5G. Several research opportunities are given as example for inspiration purpose. <s> BIB001 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> V. MCMTC ENABLING TECHNOLOGIES <s> URLLC is a new service category in 5G to accommodate emerging services and applications having stringent latency and reliability requirements. In order to support URLLC, there should be both evolutionary and revolutionary changes in the air interface named 5G NR. In this article, we provide an up-to-date overview of URLLC with an emphasis on the physical layer challenges and solutions in 5G NR downlink. We highlight key requirements of URLLC and then elaborate the physical layer issues and enabling technologies including packet and frame structure, scheduling schemes, and reliability improvement techniques, which have been discussed in the 3GPP Release 15 standardization. <s> BIB002

The key performance requirements of mcMTC discussed above affect the design choices of every component in the communication link and the optimization process within the whole protocol stack. Thus, for enabling these requirements, the entire system needs to be redesigned; an advanced channel estimation technique, restructuring packets, frames, backhaul, storage, advanced error control coding, control signaling and corresponding multiplexing scheme must be deployed, and an advanced scheme for channel coding that is suitable for the short packet transmission should be used BIB002 . In addition, both the data and control planes may require significant enhancements and new technical solutions can be given from both the radio interface and network architecture aspects. These fundamental design changes promise benefits that contribute to realizing higher reliability, reducing latency, and improving availability , BIB001 . Below, we are going to review the enabling technologies of mcMTC towards 5G, such as air-interface waveforms, network architecture, frame structure, control channel design, spatial diversity, proactive packet drop, and coding schemes. Figure 6 depicts the mcMTC key requirements and its main enablers. X-axis denotes latency in ms while Y-axis denotes reliability from 10 −12 to 0 BLER. It shows that the ultra-reliable communications have a restrict condition in reliability between 10 −12 and 10 −8 , while they have a release latency condition goes to 100ms. In contrast, low-latency communications have a restrict condition in latency between 0.1ms and 1ms, while they have a release condition in reliability goes to less than 10 −2 BLER. However, the mcMTC requires restrict conditions in both latency and reliability which go from 0.1 to 1 ms and from 10 −12 to 10 −9 BLER, respectively.

Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> A. AIR INTERFACE ENABLING TECHNOLOGIES <s> The standardization of the next generation 5G radio access technology has just started in 3GPP with the ambition of making it commercially available by 2020. There are a number of features that are unique for 5G radio access compared to the previous generations such as a wide range of carrier frequencies and deployment options, diverse use cases with very different user requirements, small-size base stations, self-backhaul, massive MIMO, and large channel bandwidths. In this article, we propose a flexible physical layer for the NR to meet the 5G requirements. A symmetric physical layer design with OFDM is proposed for all link types, including uplink, downlink, device-to-device, and backhaul. A scalable OFDM waveform is proposed to handle the wide range of carrier frequencies and deployments. <s> BIB001 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> A. AIR INTERFACE ENABLING TECHNOLOGIES <s> URLLC is a new service category in 5G to accommodate emerging services and applications having stringent latency and reliability requirements. In order to support URLLC, there should be both evolutionary and revolutionary changes in the air interface named 5G NR. In this article, we provide an up-to-date overview of URLLC with an emphasis on the physical layer challenges and solutions in 5G NR downlink. We highlight key requirements of URLLC and then elaborate the physical layer issues and enabling technologies including packet and frame structure, scheduling schemes, and reliability improvement techniques, which have been discussed in the 3GPP Release 15 standardization. <s> BIB002

It is not possible to fulfill the challenging requirements of mcMTC with minor modifications of recent radio access technologies. For that, a new air interface is required to be designed in 5G to deal with heterogeneous traffic types and to fulfill mcMTC key requirements. These requirements may vary depending on the selected use case and application. Therefore, the new air interface and system design should be scalable enough to efficiently use the available network resources. To address the key requirements discussed above, several designs have been introduced for 5G air interface. The 3GPP technical support presented in shows that the NR access technology has been defined which in general has to add entirely new features providing a customized connection to any device from different smart devices platforms and bring the backward compatibility. The early features of this NR already standardized by 3GPP in July 2018 and the complete collection of features are to be finalized by the year 2020 (IMT2020) BIB002 . Several features are specific to NR in comparison with LTE shown in Table 3 . These include mcMTC features, the broad range of carrier frequencies, several deployment options, configuration mechanisms, such as flexible waveform and protocols, adaptive frame structure, coding and modulation, multiple access schemes, and various use cases which is either machine-centric or human-centric and beyond eMBB. In addition, NR should support dynamic resource sharing between different latency and reliability requirements for both eMBB and mcMTC in downlink. However, the dynamic resource sharing between mcMTC and eMBB is supported by transmitting mcMTC scheduled traffic where mcMTC transmission may occur in resources scheduled for ongoing eMBB traffic. Downlink dynamic resources sharing between eMBB and mcMTC is enabled without pre-emption by scheduling the eMBB and mcMTC services on non-overlapping time/frequency resources . As shown in Figure 7 , NR and LTE are integral parts of 5G radio access. LTE is predicted to function below 6 GHz frequencies, while NR is expected to function from sub-1 GHz up to 100 GHz. The aggregation of NR and LTE traffic shall be enabled by tight integrations BIB001 . Huawei proposed that NR should be built upon software-defined flexible air interface and radio access virtualization . In terms of air interface, it should be optimized in a specific way supporting various mcMTC use cases of applications without contradiction with other operating modes (i.e. eMBB and mMTC). In terms of virtualization of radio access, it encloses coordination and self-organization algorithms, which employ the protocols, features, and interfaces evading the limitations of the geographic ''cell'' construct. To reach 5G era, the community of mobile communications has witnessed plentiful waveform proposals for NR and has come to the fact that there is no waveform that introduces all the required advantages. Several waveform proposals have been presented and the trend has been to tweaking Orthogonal Frequency-Division Multiplexing (OFDM) in any possible way, such as pulse shaping, wise filtering of sub-carriers, filtering of groups of sub-carriers, dropping cyclic-Prefix (CP), replacing CP with nulls or with other sequences, and allowing successive symbols to overlap in time. Even though OFDM has many advantages and it has been a great success, there exist several ideas for new 5G waveforms, which could add other advantages to the new cellular system under specific circumstances and conditions. However, multi-carrier waveforms are either variations of CP-OFDM (e.g., Unique Word OFDM, Pulse Shaped OFDM, Windowed-OFDM, Universally Filtered-OFDM, Filtered-OFDM) or super cases of OFDM (e.g., OFDM becomes a specialized case of a more complicated waveform, such as Filter-Bank Multi-Carrier waveforms). In addition to these discussed waveforms, others are being considered to be used with 5G. OFDM is considered as a baseline for up to 30 GHz. Unfortunately, a major drawback of OFDM systems is their large peak to average power ratio (PAPR) .

Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> 1) CP-OFDM <s> The fifth generation (5G) of cellular networks is starting to be defined to meet the wireless connectivity demands for 2020 and beyond. One area that is considered increasingly important is the capability to provide ultra-reliable and low-latency communication, to enable e.g., new mission-critical machine-type communication use cases. One such example with extremely demanding requirements is the industrial automation with a need for ultra-low latency with a high degree of determinism. In this paper, we discuss the feasibility, requirements and design challenges of an OFDM based 5G radio interface that is suitable for mission-critical MTC. The discussion is further accompanied with system-level performance evaluations that are carried out for a factory hall-wide automation scenario with two different floor layouts. <s> BIB001 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> 1) CP-OFDM <s> The standardization of the next generation 5G radio access technology has just started in 3GPP with the ambition of making it commercially available by 2020. There are a number of features that are unique for 5G radio access compared to the previous generations such as a wide range of carrier frequencies and deployment options, diverse use cases with very different user requirements, small-size base stations, self-backhaul, massive MIMO, and large channel bandwidths. In this article, we propose a flexible physical layer for the NR to meet the 5G requirements. A symmetric physical layer design with OFDM is proposed for all link types, including uplink, downlink, device-to-device, and backhaul. A scalable OFDM waveform is proposed to handle the wide range of carrier frequencies and deployments. <s> BIB002

Ericsson's researchers have performed a thorough assessment of waveforms. They have chosen an OFDM based air interface in which inter-symbol interference can be kept minimal by setting CP-OFDM symbol longer than the channel delay spread. They have chosen the CP-OFDM as the most appropriate candidate for NR for many reasons listed in BIB002 . Two drawbacks exist in OFDM similar to all other multi-carrier waveforms which are high PAPR and less frequency localization. However, there are simple techniques that are well-established to improve frequency localization (i.e. windowing) and to reduce PAPR (e.g. with companding and clipping). These techniques could be used easily with CP-OFDM at the transmitter in a receiver agnostic way. However, Ericsson is the company, which realizes earlier that CP-OFDM has been the best potential for NR among the available candidates. However, Ericsson researchers have proposed this waveform assessment through 3GPP on April 2016 (the first 3GPP RAN 1 meeting on 5G New Radio) and have introduced CP-OFDM for 5G NR. Ericsson has suggested some modifications to OFDM including reduced TTI and shorter OFDM symbol durations that enable fast and efficient data transmission and thus satisfy the requirement of mcMTC. Since the LTE symbol length could be too long in relation to the envisioned air interface requirements (e.g., maximal TTI is assumed to be 0.1 ms for the time-critical factory automation applications BIB001 ), both the OFDM symbol length and TTI need to be scaled down for mcMTC. Reduced OFDM symbol length, error floor-free data coding, and higher-order antenna diversity are several examples of major technical enablers identified for the realization of mcMTC.

Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> 4) SCMA <s> Fifth generation (5G) wireless networks are expected to support very diverse applications and terminals. Massive connectivity with a large number of devices is an important requirement for 5G networks. Current LTE system is not able to efficiently support massive connectivity, especially on the uplink (UL). Among the issues that arise due to massive connectivity is the cost of signaling overhead and latency. In this paper, an uplink contention-based sparse code multiple access (SCMA) design is proposed as a solution. First, the system design aspects of the proposed multiple-access scheme are described. The SCMA parameters can be adjusted to provide different levels of overloading, thus suitable to meet the diverse traffic connectivity requirements. In addition, the system-level evaluations of a small packet application scenario are provided for contention-based UL SCMA. SCMA is compared to OFDMA in terms of connectivity and drop rate under a tight latency requirement. The simulation results demonstrate that contention-based SCMA can provide around 2.8 times gain over contention-based OFDMA in terms of supported active users. The uplink contention-based SCMA scheme can be a promising technology for 5G wireless networks for data transmission with low signaling overhead, low delay, and support of massive connectivity. <s> BIB001 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> 4) SCMA <s> Compared with existing systems, fifth generation (5G) mobile systems are expected to support diverse services and massive connectivity. Thus, multiple access schemes play a critical role. Recently, a codebook-based non-orthogonal access mode called sparse code multiple access (SCMA) has been proved to enable massive connectivity by allowing SCMA layer overloading. Message Passing Algorithm (MPA) can iteratively detect the multiplexed SCMA codewords. However, reducing MPA computation complexity is still a problem since the 5G ultra low latency and energy consumption targets. As the conditional channel probability (CCP) calculation takes up 60% computation of logarithm-domain MPA, reducing the CCP calculation is the tipping point of MPA optimization. In this paper, we proposed a dynamic shrunk square searching (DSSS) algorithm, based on the signal uncertainty theory, to dynamically shrink the state space before the CCP calculation. As the DSSS exploits both the structure of state space and noise characteristic, it could alleviate unnecessary CCP calculation while maintaining the near optimal decoding performance in block error (BLER) sense. <s> BIB002

Sparse code multiple access (SCMA) is a new frequency domain non-orthogonal multiple-access technique. SCMA is favorable for massive connectivity due to its tolerance to overloading signals. In addition, SCMA uses blind detection technique; therefore, it can be considered as an enabler for grant-free multiple access which can effectively terminates the latency and signaling in the request-grant dynamic scheduling schemes BIB001 - BIB002 . As a result, it is anticipated that the final outcome for 5G waveforms may have an adaptive solution utilizing the optimal waveform for any introduced situation. 5G is being investigated and explored. For that, there is a good opportunity to consider the optimal waveforms for the 5G system, which shall be employed until 2040 at least. In March 2016, 3GPP has agreed to study several features of NR taking into account OFDM, unless significant gains could be presented by other waveforms . In August 2016, 3GPP has agreed to use CP-OFDM for both uplink and downlink in NR. Thus, CP-OFDM takes even a greater role in 5G NR compared with 4G LTE. 3GPP has settled on CP-OFDM and most proposals . Table 4 shows the different waveforms used by all wireless network generations from G1 to the upcoming 5G.

Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> 1) NETWORK SLICING <s> 5G is envisioned to support unprecedented diverse applications and services with extremely heterogeneous performance requirements, i.e., mission critical IoT communication, massive machine-type communication and Gigabit mobile connectivity. This imposes enormous challenges to fulfil the key performance requirements, in particular, mission critical IoT communication, which calls for a dramatic paradigm change in 5G. This paper presents vision and challenges of mission critical IoT scenarios and the enabling technologies in 5G. Several research opportunities are given as example for inspiration purpose. <s> BIB001 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> 1) NETWORK SLICING <s> The grand objective of 5G wireless technology is to support three generic services with vastly heterogeneous requirements: enhanced mobile broadband (eMBB), massive machine-type communications (mMTC), and ultra-reliable low-latency communications (URLLC). Service heterogeneity can be accommodated by network slicing, through which each service is allocated resources to provide performance guarantees and isolation from the other services. Slicing of the Radio Access Network (RAN) is typically done by means of orthogonal resource allocation among the services. This work studies the potential advantages of allowing for non-orthogonal sharing of RAN resources in uplink communications from a set of eMBB, mMTC and URLLC devices to a common base station. The approach is referred to as Heterogeneous Non-Orthogonal Multiple Access (H-NOMA), in contrast to the conventional NOMA techniques that involve users with homogeneous requirements and hence can be investigated through a standard multiple access channel. The study devises a communication-theoretic model that accounts for the heterogeneous requirements and characteristics of the three services. The concept of reliability diversity is introduced as a design principle that leverages the different reliability requirements across the services in order to ensure performance guarantees with non-orthogonal RAN slicing. This study reveals that H-NOMA can lead, in some regimes, to significant gains in terms of performance trade-offs among the three generic services as compared to orthogonal slicing. <s> BIB002

Network Slicing is one of the main concepts by which the challenges are addressed by enabling a higher degree of flexibility in the network such that each slice could be customized to meet the requirements of a certain use case in order to optimize the utilization of network resources BIB001 , BIB002 . Network Slicing is a new concept to allow differentiated treatment depending on requirements of each customer . The 5G network shall be designed in a way making it flexible enough for creating a virtual instance of an entire network for every different use case. Various customized virtual networks shall exist simultaneously and they shall not interfere with each other. As an example, a customized virtual network, which requires a very high throughput for video streaming, can co-exist with a customized virtual network for ultra-low latency autonomous vehicle control. Figure 10 depicts the slicing concept in 5G where it was not in 4G.

Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> 5) CLOUDRAN ARCHITECTURE <s> OpenFlow (OF) is one of the most widely used protocols for controller-to-switch communication in a software defined network (SDN). Performance analysis of OF-based SDN using analytical models is both highly desirable and challenging. There already exists a very elegant analytical model based on M/M/1 queues to estimate the packet sojourn time and probability of lost packets for the case in which a controller is responsible for only a single node in the data plane. However the literature falls short when it comes to the multiple node case, i.e. when there is more than one node in the data plane. In this work we propose a model to address this challenge by approximating the data plane as an open Jackson network with the controller also modeled as an M/M/1 queue. The model is then used to evaluate the system in the light of some of the metrics, such as; how much time a packet spends on average in an OF-based network and how much data we can pump into the network given the average delay requirements. Finally the PDF and the CDF of the time spent by the packet in an OF-based SDN for a given path is derived. <s> BIB001 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> 5) CLOUDRAN ARCHITECTURE <s> Next generation mobile networks not only envision enhancing the traditional MBB use case but also aim to meet the requirements of new use cases, such as the IoT. This article focuses on latency critical IoT applications and analyzes their requirements. We discuss the design challenges and propose solutions for the radio interface and network architecture to fulfill these requirements, which mainly benefit from flexibility and service-centric approaches. The article also discusses new business opportunities through IoT connectivity enabled by future networks. <s> BIB002 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> 5) CLOUDRAN ARCHITECTURE <s> Flexible network architecture is envisioned as one of the properties of the next generation mobile networks, and is also foreseen to significantly contribute to lowering latency, and addressing Ultra-Reliable Low Latency Communications (URLLC). In this paper, we study the flexibility in Radio Access Network (RAN) configuration in terms of splitting the radio and baseband functionalities between central cloud and distributed entities, and the impact of this flexibility on delivery of URLLC. Three different functionality splits have been implemented in cloud- RAN, and the impact of each of these splits on communication latency and jitter is examined. We modeled traffic based on the 3GPP traffic models and, for completeness, in addition to URLLC, the two cases of massive machine-type communications (mMTC) and enhanced Mobile Broadband (eMBB) models are also implemented. Hence, thorough analyses are performed and recommendation for split point between central cloud and distributed radio units are discussed. Index Terms-5G; Cloud-RAN; functionality split; URLLC; eMBB; mMTC. <s> BIB003

CloudRAN architecture is envisioned as one of the properties of 5G to enable a flexible, and scalable architecture that can be adjusted to the needs of several use cases which run concurrently on the same infrastructure. This pattern also foreseen to significantly contribute to addressing mcMTC requirements. Providing flexibility in RAN configuration in terms of splitting the radio and baseband functionalities between central cloud and distributed entities will contribute delivery of mcMTC. In particular, latency critical MTC use cases can benefit from local computational power provided by applications running in the MEC since this reduces the physical and virtual communication distance BIB002 . However, network functions in 5G will increasingly be deployed based on cloud SDN and NFV. SDN is about the separation of the network control traffic (control plane) and the user specific traffic (data plane). Meanwhile, NFV is about virtualizing network functions and the functions that can run on a range of standard hardware . CloudRAN, simplifies scaling and management of network infrastructure such as deep packet inspection engines and firewalls BIB003 . Examples of how flow-level modeling can be applied on SDN/NFV architectures may be found in BIB001 . However, SDN and NFV are new technology but not included in 3GPP technical specifications yet.

Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> D. PACKET AND FRAME STRUCTURE <s> Wireless 5G systems will not only be "4G, but faster". One of the novel features discussed in relation to 5G is Ultra-Reliable Communication (URC), an operation mode not present in today's wireless systems. URC refers to provision of certain level of communication service almost 100 % of the time. Example URC applications include reliable cloud connectivity, critical connections for industrial automation and reliable wireless coordination among vehicles. This paper puts forward a systematic view on URC in 5G wireless systems. It starts by analyzing the fundamental mechanisms that constitute a wireless connection and concludes that one of the key steps towards enabling URC is revision of the methods for encoding control information (metadata) and data. It introduces the key concept of Reliable Service Composition, where a service is designed to adapt its requirements to the level of reliability that can be attained. The problem of URC is analyzed across two different dimensions. The first dimension is the type of URC problem that is defined based on the time frame used to measure the reliability of the packet transmission. Two types of URC problems are identified: long-term URC (URC-L) and short-term URC (URC-S). The second dimension is represented by the type of reliability impairment that can affect the communication reliability in a given scenario. The main objective of this paper is to create the context for defining and solving the new engineering problems posed by URC in 5G. <s> BIB001 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> D. PACKET AND FRAME STRUCTURE <s> A 5G frame structure designed for efficient support of users with highly diverse service requirements is proposed. It includes support for mobile broadband data, mission-critical communication, and massive machine communication. The solution encompasses flexible multiplexing of users on a shared channel with dynamic adjustment of the transmission time interval in coherence with the service requirements per link. This allows optimizing the fundamental tradeoffs between spectral efficiency, latency, and reliability for each link and service flow. The frame structure is based on in-resource physical layer control signaling that follows the corresponding data transmission for each individual user. Comparison against the corresponding LTE design choices shows attractive benefits. <s> BIB002 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> D. PACKET AND FRAME STRUCTURE <s> URLLC is a new service category in 5G to accommodate emerging services and applications having stringent latency and reliability requirements. In order to support URLLC, there should be both evolutionary and revolutionary changes in the air interface named 5G NR. In this article, we provide an up-to-date overview of URLLC with an emphasis on the physical layer challenges and solutions in 5G NR downlink. We highlight key requirements of URLLC and then elaborate the physical layer issues and enabling technologies including packet and frame structure, scheduling schemes, and reliability improvement techniques, which have been discussed in the 3GPP Release 15 standardization. <s> BIB003

One of the most important objectives of 5G systems is supporting different packet lengths (1-10 MB) and various symbol numerologies (e.g., sampling rate, subcarrier spacing, symbol duration) in order to accommodate various deployment scenarios and service requirements BIB003 . The main aim of the design of mcMTC packet is minimizing the processing latency, which includes time for acquisition of channel information, receiving data symbols, decoding data packets, checking the existence of error, and scheduling information. The frame structure is dependent on in-resource PHY layer control signaling, which follows the corresponding data transmission for each user. Comparison against the corresponding LTE design choices shows attractive benefits. In addition, enhancement of HARQ retransmissions and Flexible design entail a flexible duration of the ACK/NACK duration and configurability per user BIB002 . In LTE the 1ms TTI and 8ms waiting time at every retransmission results in degrading the E2E latency to more than 20 ms . Hence, existing LTE frames cannot support the mcMTC service due to the structure of LTE resource blocks of 1ms TTI and 14 symbols size. Therefore, 5G requires a shorter frame and faster processing time at retransmissions to produce a lower latency. In Figure 11 , the size of the 5G resource block is 3 symbols with 72ms TTI each, while in LTE the size of the resource block is 14 symbols and 1ms TTI. Therefore, the packet transmission time should be in the order of hundreds of microseconds to meet this stringent latency constraint . In 4G systems, a rectangular-shaped packet structure is used to make an efficient utilization of the spectrum under time-frequency fading channel. In mcMTC systems, on the other hand, a non-rectangular packet is favored because employing this structure shall save time to receive the packet data. To reduce the latency further, BIB003 suggested that the three components of a packet (pilot, control, data part) should be grouped together and afterward transmitted sequentially. By performing this, data detection, control channel decoding, and pipelined processing of the channel estimation become possible. Furthermore, in the control part, employing a simple fast decoding scheme is preferred to relying on a time-consuming process that consists of channel decoding and blind searching. As suggested in BIB001 , the header and data should be enclosed into a single packet to increase the probability of successfully receiving the header and the data. This idea is very relevant to services that use short messages, such as mcMTC. This is due to the header size and data size that are of comparable size (i.e. the header cannot be sent sub-optimally by using repetition coding and a low header data rate). Popovski's solution could work in practice, but the paper does not provide simulation results or physical results to confirm the proposed idea. It also does not address the situation where the application is located at the cell edge, which may have little to no mobile Internet reception.

Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> F. COMMUNICATION DIVERSITY <s> Fifth generation wireless networks are currently being developed to handle a wide range of new use cases. One important emerging area is ultra-reliable communication with guaranteed low latencies well beyond what current wireless technologies can provide. In this paper, we explore the viability of using wireless communication for low-latency, high-reliability communication in an example scenario of factory automation, and outline important design choices for such a system. We show that it is possible to achieve very low error rates and latencies over a radio channel, also when considering fast fading signal and interference, channel estimation errors, and antenna correlation. The most important tool to ensure high reliability is diversity, and low latency is achieved by using short transmission intervals without retransmissions, which, however, introduces a natural restriction on coverage area. <s> BIB001 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> F. COMMUNICATION DIVERSITY <s> The fifth generation (5G) of cellular networks is starting to be defined to meet the wireless connectivity demands for 2020 and beyond. One area that is considered increasingly important is the capability to provide ultra-reliable and low-latency communication, to enable e.g., new mission-critical machine-type communication use cases. One such example with extremely demanding requirements is the industrial automation with a need for ultra-low latency with a high degree of determinism. In this paper, we discuss the feasibility, requirements and design challenges of an OFDM based 5G radio interface that is suitable for mission-critical MTC. The discussion is further accompanied with system-level performance evaluations that are carried out for a factory hall-wide automation scenario with two different floor layouts. <s> BIB002 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> F. COMMUNICATION DIVERSITY <s> Factory automation is one of the challenging use cases that the fifth generation, 5G, networks are expected to support. It involves mission-critical machine-type communications, MTC, with requirements of extreme low-latency and ultra-reliable communication to enable real-time control of automation processes in manufacturing facilities. In this paper, we discuss the deployment strategies for the 5G mission-critical MTC solution designed to meet the needs of factory automation applications. The paper analyzes the coverage and capacity aspects based on a series of system-level evaluations considering both noise-limited and interference- limited operations. It further analyzes the related trade-offs to provide insights on the network deployment strategies for a realistic factory scenario. <s> BIB003

In wireless communications, the most difficult case regarding performing high reliability is the Rayleigh fading channel. This is due to large fading dips in this channel. To improve the reliability and availability of signal detection and decoding, a higher-order diversity is essential. Therefore, diversity is considered to be one of the most significant techniques to achieve high reliability in a fading environment. In general, diversity is the capability to use the three variations which are namely time, frequency and/or space to exploit the channel diversity gains. Such that, the latency bounds dictated by the time-critical mcMTC applications limit the number of possible retransmissions. Moreover, the exploitation of frequency diversity would be costly due to the need for frequency resources having uncorrelated channel coefficients. Therefore, antenna diversity is identified as the major technique to overcome uncertainty and fading dips in the channel BIB001 and it is found beneficial in mcMTC deployments BIB002 , BIB003 . In order to fully occupy the benefits of frequency diversity over the frequency domain that have uncorrelated channel coefficients, the coded bits are to be mapped on the resources. Additionally, combining multiple diversity techniques such as frequency, antenna, and various forms of spatial diversity can offer extra gains to approach the high reliability target within stringent latency constraints . Such that, antenna diversity uses more antennas to improve the quality and reliability of a wireless link, while frequency diversity allows receiving radio signals over multiple channels (different frequencies) or wide radio channel (wide frequency band) to reduce the effects of radio signal distortions such as signal fading.

Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> 1) CONVOLUTIONAL CODING SCHEME <s> Future generation of wireless networks, i.e. 5G, is envisioned to support several new use-cases demanding transmission reliability and latency that cannot be achieved by the current cellular networks such as long-term evolution (LTE). This paper looks at different design aspects of the control channel(s) to support ultra-reliable low-latency communication considering factory automation as an example scenario. In particular, we show that a fairly balanced design for both the uplink and the downlink control channels can be made given an appropriate selection of modulation, coding, diversity scheme, and time/frequency resources. By means of link-level simulations, we also show that the proposed control channel design supports a block-error rate of 10-9 under Rayleigh fading conditions at a signal-to-interference-plus-noise ratio comparable to that supported by current 4G systems (e.g. LTE). Furthermore, a radio frame structure is proposed to support the user plane end-to-end latency of 1 ms. <s> BIB001 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> 1) CONVOLUTIONAL CODING SCHEME <s> Fifth generation wireless networks are currently being developed to handle a wide range of new use cases. One important emerging area is ultra-reliable communication with guaranteed low latencies well beyond what current wireless technologies can provide. In this paper, we explore the viability of using wireless communication for low-latency, high-reliability communication in an example scenario of factory automation, and outline important design choices for such a system. We show that it is possible to achieve very low error rates and latencies over a radio channel, also when considering fast fading signal and interference, channel estimation errors, and antenna correlation. The most important tool to ensure high reliability is diversity, and low latency is achieved by using short transmission intervals without retransmissions, which, however, introduces a natural restriction on coverage area. <s> BIB002 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> 1) CONVOLUTIONAL CODING SCHEME <s> Most of the recent advances in the design of high-speed wireless systems are based on information-theoretic principles that demonstrate how to efficiently transmit long data packets. However, the upcoming wireless systems, notably the fifth-generation (5G) system, will need to support novel traffic types that use short packets. For example, short packets represent the most common form of traffic generated by sensors and other devices involved in machine-to-machine (M2M) communications. Furthermore, there are emerging applications in which small packets are expected to carry critical information that should be received with low latency and ultrahigh reliability. Current wireless systems are not designed to support short-packet transmissions. For example, the design of current systems relies on the assumption that the metadata (control information) is of negligible size compared to the actual information payload. Hence, transmitting metadata using heuristic methods does not affect the overall system performance. However, when the packets are short, metadata may be of the same size as the payload, and the conventional methods to transmit it may be highly suboptimal. In this paper, we review recent advances in information theory, which provide the theoretical principles that govern the transmission of short packets. We then apply these principles to three exemplary scenarios (the two-way channel, the downlink broadcast channel, and the uplink random access channel), thereby illustrating how the transmission of control information can be optimized when the packets are short. The insights brought by these examples suggest that new principles are needed for the design of wireless protocols supporting short packets. These principles will have a direct impact on the system design. <s> BIB003

Convolutional code is selected as the preferred Forward Error Correction (FEC) scheme in mcMTC use cases due to many reasons listed in BIB001 . In addition, the decoder is able to process the code block whilst it is being received, and thus it is able to obtain decoded bits with a very short delay. However, for better performance and manageable decoding complexity, the block codes are favorable for control channels which have short blocks lower than 10 bits. Therefore, redesign of PHY channels allows early channel estimation usage of convolutional codes (e.g., for data channels) and block codes (e.g., for control channels) , BIB002 . However, convolutional codes perform similarly to Turbo and LDPC codes for block lengths, which are up to a few hundred bits. On the contrary of convolutional codes, Turbo and LDPC codes might in specific configurations have an error floor, which makes these codes less efficient when BLER reaches very low levels. Considering latency, convolutional code decoding has a shorter delay than the iterative decoder, which is usually employed in Turbo and LDPC decoding. This is partially because of lower decoding complexity. 2) TURBO CODING SCHEME Turbo codes have been used as FEC for data in several modern communication systems, such as LTE. Turbo code encoder is built using a parallel concatenation of two recursive systematic convolutional codes and the associated decoder, using a feedback decoding rule. Turbo codes have low error probability performance within a 1dB fraction from the Shannon limit and relatively low complexity which is about 10 −6 BLER . Although it is being used in LTE, it does not satisfy the performance requirements of 5G services for all the code rates and block lengths as the implementation complexity is too high for higher data rates. In addition, due to the error floor observed in this scheme. Turbo codes generally have a low encoding complexity and high decoding complexity. 3) LDPC CODING SCHEME Since LDPC codes were rediscovered in 1993. LDPC codes have attracted growing interests in both academia and industry because it can significantly improve the performance of both wired and wireless communication systems. LDPC codes are linear codes and have a sparse parity check matrix consisting low density which have relatively simple and practical decoding algorithms. The iterative nature of LDPC decoding algorithms increases the accuracy in each iteration while the number of iterations is decided based on the requirement of the application. Compared with turbo codes, LDPC codes can achieve better performance and faster decoding such that LDPC codes have an excellent ability to achieve almost 95% of the theoretical limits of channel capacity BIB003 . In addition, LDPC codes have the ability to trade-off between latency, reliability performance which make it a good candidate for future communications supporting mcMTC. However, modern LDPC decoders work with soft decision algorithms which further enhance the decoder gain. In eMBB the channel coding scheme for data is the flexible LDPC and is the same as the single channel coding scheme for all block sizes. In addition, the channel coding scheme is Polar Coding with exception to very small block lengths where repetition/block coding is preferred. However, LDPC codes are currently being used in many communication systems such as 802.11n (Wi-Fi allowing MIMO) and 802.16e (Mobile WiMAX) etc. However, LDPC coding has been selected for a control channel coding in 5G .

Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> 4) POLAR CODING SCHEME <s> The recently-discovered polar codes are widely seen as a major breakthrough in coding theory. These codes achieve the capacity of many important channels under successive cancellation decoding. Motivated by the rapid progress in the theory of polar codes, we propose a family of architectures for efficient hardware implementation of successive cancellation decoders. We show that such decoders can be implemented with O(n) processing elements and O(n) memory elements, while providing constant throughput. We also propose a technique for overlapping the decoding of several consecutive codewords, thereby achieving a significant speed-up factor. We furthermore show that successive cancellation decoding can be implemented in the logarithmic domain, thereby eliminating the multiplication and division operations and greatly reducing the complexity of each processing element. <s> BIB001 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> 4) POLAR CODING SCHEME <s> We describe a successive-cancellation list decoder for polar codes, which is a generalization of the classic successive-cancellation decoder of Arikan. In the proposed list decoder, up to L decoding paths are considered concurrently at each decoding stage. Simulation results show that the resulting performance is very close to that of a maximum-likelihood decoder, even for moderate values of L. Thus it appears that the proposed list decoder bridges the gap between successive-cancellation and maximum-likelihood decoding of polar codes. The specific list-decoding algorithm that achieves this performance doubles the number of decoding paths at each decoding step, and then uses a pruning procedure to discard all but the L “best” paths. In order to implement this algorithm, we introduce a natural pruning criterion that can be easily evaluated. Nevertheless, straightforward implementation still requires O(L · n2) time, which is in stark contrast with the O(n log n) complexity of the original successive-cancellation decoder. We utilize the structure of polar codes to overcome this problem. Specifically, we devise an efficient, numerically stable, implementation taking only O(L · n log n) time and O(L · n) space. <s> BIB002 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> 4) POLAR CODING SCHEME <s> Flexibility is one mandatory aspect of channel coding in modern wireless communication systems. Among other things, the channel decoder has to support several code lengths and code rates. This need for flexibility applies to polar codes that are considered for control channels in the future 5G standard. This paper presents a new generic and flexible implementation of a software Successive Cancellation List (SCL) decoder. A large set of parameters can be fine-tuned dynamically without re-compiling the software source code: the code length, the code rate, the frozen bits set, the puncturing patterns, the cyclic redundancy check, the list size, the type of decoding algorithm, the tree-pruning strategy and the data quantization. This generic and flexible SCL decoder enables to explore tradeoffs between throughput, latency and decoding performance. Several optimizations are proposed to achieve a competitive decoding speed despite the constraints induced by the genericity and the flexibility. The resulting polar list decoder is about 4 times faster than a generic software decoder and only 2 times slower than a non-flexible unrolled decoder. Thanks to the flexibility of the decoder, the fully adaptive SCL algorithm can be easily implemented and achieves higher throughput than any other similar decoder in the literature (up to 425 Mb/s on a single processor core for N = 2048 and K = 1723 at 4.5 dB). <s> BIB003

In coding theory, polar codes are considered as a big advancement. When the code block size is large enough, polar codes are able to perform Shannon capacity with a simple encoder and a simple Successive Cancellation (SC) decoder. Polar codes have raised much interest so that several research works have been mainly conducted on code design and decoding algorithm BIB001 . Polar codes outperform all the codes, which are currently used in the 4G systems, specifically for short code length BIB002 . Hence, polar codes are treated as great potentials for the FEC module in 5G air interface design . Many variants of Polar codes have been proposed to improve the performance of Polar codes, some concatenated coding and combined decoding schemes are proposed. Such as CRC-concatenated Polar codes (CA-Polar Code) with single-parity-check code concatenation and multi CRC-concatenation Parity-check concatenated Polar code (PC-Polar Code). Furthermore, the performance of Polar Codes keeps improving along with increasing list size of the software Successive Cancellation List (SCL) decoder. The flexibility of SCL decoder enables to explore trade-offs between throughput, latency decoding performance BIB003 .

Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> 5) COMBINATION OF POLAR, SC, AND CRC CODING SCHEME <s> This paper investigates the maximal channel coding rate achievable at a given blocklength and error probability. For general classes of channels new achievability and converse bounds are given, which are tighter than existing bounds for wide ranges of parameters of interest, and lead to tight approximations of the maximal achievable rate for blocklengths n as short as 100. It is also shown analytically that the maximal rate achievable with error probability ? isclosely approximated by C - ?(V/n) Q-1(?) where C is the capacity, V is a characteristic of the channel referred to as channel dispersion , and Q is the complementary Gaussian cumulative distribution function. <s> BIB001 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> 5) COMBINATION OF POLAR, SC, AND CRC CODING SCHEME <s> In this paper, we propose a capacity-approaching analog fountain code (AFC) for wireless channels. In AFC, the number of generated coded symbols is potentially limitless. In contrast to the conventional binary rateless codes, each coded symbol in AFC is a real-valued symbol, generated as a weighted sum of d randomly selected information bits, where d and the weight coefficients are randomly selected from predefined probability mass functions. The coded symbols are then directly transmitted through wireless channels. We analyze the error probability of AFC and design the weight set to minimize the error probability. Simulation results show that AFC achieves the capacity of the Gaussian channel in a wide range of signal to noise ratio (SNR). <s> BIB002 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> 5) COMBINATION OF POLAR, SC, AND CRC CODING SCHEME <s> 5G is envisioned to support unprecedented diverse applications and services with extremely heterogeneous performance requirements, i.e., mission critical IoT communication, massive machine-type communication and Gigabit mobile connectivity. This imposes enormous challenges to fulfil the key performance requirements, in particular, mission critical IoT communication, which calls for a dramatic paradigm change in 5G. This paper presents vision and challenges of mission critical IoT scenarios and the enabling technologies in 5G. Several research opportunities are given as example for inspiration purpose. <s> BIB003 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> 5) COMBINATION OF POLAR, SC, AND CRC CODING SCHEME <s> This paper investigates block error rate (BLER) performance and computational complexity of candidate channel coding schemes for ultra-reliable low latency communication (URLLC) in 5G. The considered candidates are the same as those identified in 3GPP: turbo, LDPC, polar, and convolutional codes. Details of code constructions and decoding algorithms are provided with computational complexity analysis. Code construction parameters, number of iterations, and list sizes are selected to provide a fair comparison among candidate coding schemes. Simulation results on BLER are shown for several code rates and small-to-moderate block sizes. The results reveal that polar and LDPC codes outperform turbo codes for short block sizes of 40 bits, while the opposite is true for medium block sizes of 200 bits. None of the schemes is a clear winner at all considered block sizes and coding rates. Other aspects like implementation complexity, latency, and flexibility will also be important when deciding the URLLC coding scheme. <s> BIB004 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> 5) COMBINATION OF POLAR, SC, AND CRC CODING SCHEME <s> Most of the recent advances in the design of high-speed wireless systems are based on information-theoretic principles that demonstrate how to efficiently transmit long data packets. However, the upcoming wireless systems, notably the fifth-generation (5G) system, will need to support novel traffic types that use short packets. For example, short packets represent the most common form of traffic generated by sensors and other devices involved in machine-to-machine (M2M) communications. Furthermore, there are emerging applications in which small packets are expected to carry critical information that should be received with low latency and ultrahigh reliability. Current wireless systems are not designed to support short-packet transmissions. For example, the design of current systems relies on the assumption that the metadata (control information) is of negligible size compared to the actual information payload. Hence, transmitting metadata using heuristic methods does not affect the overall system performance. However, when the packets are short, metadata may be of the same size as the payload, and the conventional methods to transmit it may be highly suboptimal. In this paper, we review recent advances in information theory, which provide the theoretical principles that govern the transmission of short packets. We then apply these principles to three exemplary scenarios (the two-way channel, the downlink broadcast channel, and the uplink random access channel), thereby illustrating how the transmission of control information can be optimized when the packets are short. The insights brought by these examples suggest that new principles are needed for the design of wireless protocols supporting short packets. These principles will have a direct impact on the system design. <s> BIB005 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> 5) COMBINATION OF POLAR, SC, AND CRC CODING SCHEME <s> URLLC is a new service category in 5G to accommodate emerging services and applications having stringent latency and reliability requirements. In order to support URLLC, there should be both evolutionary and revolutionary changes in the air interface named 5G NR. In this article, we provide an up-to-date overview of URLLC with an emphasis on the physical layer challenges and solutions in 5G NR downlink. We highlight key requirements of URLLC and then elaborate the physical layer issues and enabling technologies including packet and frame structure, scheduling schemes, and reliability improvement techniques, which have been discussed in the 3GPP Release 15 standardization. <s> BIB006 </s> Mission-Critical Machine-Type Communication: An Overview and Perspectives Towards 5G <s> 5) COMBINATION OF POLAR, SC, AND CRC CODING SCHEME <s> The fifth-generation cellular mobile networks are expected to support mission critical ultra-reliable low latency communication (uRLLC) services in addition to the enhanced mobile broadband applications. This article first introduces three emerging mission critical applications of uRLLC and identifies their requirements on end-to-end latency and reliability. We then investigate the various sources of end-to-end delay of current wireless networks by taking the 4G Long Term Evolution (LTE) as an example. Subsequently, we propose and evaluate several techniques to reduce the end-to-end latency from the perspectives of error control coding, signal processing, and radio resource management. We also briefly discuss other network design approaches with the potential for further latency reduction. <s> BIB007

It is essential that establishing a very low latency could be gained by using a simple encoder and fast decoding algorithms. Several performance simulations show that the concatenation between Polar codes and CRC and an adaptive SC decoder are able to overcome turbo/LDPC codes for short and moderate code block sizes . The SC decoding algorithm is among the most important decoding algorithms and it is able to perform same as the optimal maximum-likelihood decoding algorithm with a list size of 32 for moderate code block sizes BIB004 . Reference BIB003 suggested that an advanced error control coding such as a concatenation of Polar codes with CRC codes could be applied to improve the reliability and promises to overcome turbo or LDPC codes. Also, reliability should be enhanced for both the control and data parts. For the control part, a new strategy, which does not depend on the CRC and channel decoding is required. For the data part, an advanced channel coding scheme that functions with the short-packet regime could be used (i.e., polar code) BIB006 . The study in BIB001 provides a method analyzing of channel capacity with finite block length called a Polyansky-Poor-Verdu (PPV). PPV has provided the trade-offs between delays, throughput, and reliability on Gaussian channels and fixed rate block codes, by introducing a new fundamental parameter called ''channel dispersion''; this analysis shows that there is a severe capacity loss at short block-lengths. There are no known codes that achieve the PPV limit. LDPC codes and polar codes have been reported to achieve almost 95% of the PPV bound at BLERs as low as 10 −7 for block lengths of a few hundred symbols BIB005 . However, their main drawback is the large decoding latency BIB007 . From the other size, Self-adaptive codes appear as a promising solution to mcMTC BIB007 . Self-adaptive codes, also known as rateless codes, can adapt the code rate to the channel variations by sending an exact amount of coded symbols needed for successful decoding. This self-adaptation does not require any CSI at the transmitter side, thus eliminating the channel estimation overhead and delay BIB002 . Table 5 summarizes the popular coding schemes and their contributions to mcMTC in 5G.

Experimental Security Analysis of Controller Software in SDNs: A Review <s> Introduction <s> This whitepaper proposes OpenFlow: a way for researchers to run experimental protocols in the networks they use every day. OpenFlow is based on an Ethernet switch, with an internal flow-table, and a standardized interface to add and remove flow entries. Our goal is to encourage networking vendors to add OpenFlow to their switch products for deployment in college campus backbones and wiring closets. We believe that OpenFlow is a pragmatic compromise: on one hand, it allows researchers to run experiments on heterogeneous switches in a uniform way at line-rate and with high port-density; while on the other hand, vendors do not need to expose the internal workings of their switches. In addition to allowing researchers to evaluate their ideas in real-world traffic settings, OpenFlow could serve as a useful campus component in proposed large-scale testbeds like GENI. Two buildings at Stanford University will soon run OpenFlow networks, using commercial Ethernet switches and routers. We will work to encourage deployment at other schools; and We encourage you to consider deploying OpenFlow in your university network too <s> BIB001 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Introduction <s> The pull of Software-Defined Networking (SDN) is magnetic. There are few in the networking community who have escaped its impact. As the benefits of network visibility and network device programmability are discussed, the question could be asked as to who exactly will benefit? Will it be the network operator or will it, in fact, be the network intruder? As SDN devices and systems hit the market, security in SDN must be raised on the agenda. This paper presents a comprehensive survey of the research relating to security in software-defined networking that has been carried out to date. Both the security enhancements to be derived from using the SDN framework and the security challenges introduced by the framework are discussed. By categorizing the existing work, a set of conclusions and proposals for future research directions are presented. <s> BIB002 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Introduction <s> This paper presents an independent comprehensive analysis of the efficiency indexes of popular open source SDN/OpenFlow controllers (NOX, POX, Beacon, Floodlight, MuL, Maestro, Ryu). The analysed indexes include performance, scalability, reliability, and security. For testing purposes we developed the new framework called hcprobe. The test bed and the methodology we used are discussed in detail so that everyone could reproduce our experiments. The result of the evaluation show that modern SDN/OpenFlow controllers are not ready to be used in production and have to be improved in order to increase all above mentioned characteristics. <s> BIB003 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Introduction <s> The idea of programmable networks has recently re-gained considerable momentum due to the emergence of the Software-Defined Networking (SDN) paradigm. SDN, often referred to as a "radical new idea in networking", promises to dramatically simplify network management and enable innovation through network programmability. This paper surveys the state-of-the-art in programmable networks with an emphasis on SDN. We provide a historic perspective of programmable networks from early ideas to recent developments. Then we present the SDN architecture and the OpenFlow standard in particular, discuss current alternatives for implementation and testing of SDN-based protocols and services, examine current and future SDN applications, and explore promising research directions based on the SDN paradigm. <s> BIB004 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Introduction <s> Software defined networking (SDN) decouples the network control and data planes. The network intelligence and state are logically centralized and the underlying network infrastructure is abstracted from applications. SDN enhances network security by means of global visibility of the network state where a conflict can be easily resolved from the logically centralized control plane. Hence, the SDN architecture empowers networks to actively monitor traffic and diagnose threats to facilitates network forensics, security policy alteration, and security service insertion. The separation of the control and data planes, however, opens security challenges, such as man-in-the middle attacks, denial of service (DoS) attacks, and saturation attacks. In this paper, we analyze security threats to application, control, and data planes of SDN. The security platforms that secure each of the planes are described followed by various security approaches for network-wide security in SDN. SDN security is analyzed according to security dimensions of the ITU-T recommendation, as well as, by the costs of security solutions. In a nutshell, this paper highlights the present and future security challenges in SDN and future directions for secure SDN. <s> BIB005 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Introduction <s> In recent years, Software-Defined Networking (SDN) has been a focus of research. As a promising network architecture, SDN will possibly replace traditional networking, as it brings promising opportunities for network management in terms of simplicity, programmability, and elasticity. While many efforts are currently being made to standardize this emerging paradigm, careful attention needs to be also paid to security at this early design stage. This paper focuses on the security aspects of SDN. We begin by discussing characteristics and standards of SDN. On the basis of these, we discuss the security features as a whole and then analyze the security threats and countermeasures in detail from three aspects, based on which part of the SDN paradigm they target, i.e., the data forwarding layer, the control layer and the application layer. Countermeasure techniques that could be used to prevent, mitigate, or recover from some of such attacks are also described, while the threats encountered when developing these defensive mechanisms are highlighted. <s> BIB006 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Introduction <s> The revolutionary concept of Software Defined Networks (SDNs) potentially provides flexible and well-managed next-generation networks. All the hype surrounding the SDNs is predominantly because of its centralized management functionality, the separation of the control plane from the data forwarding plane, and enabling innovation through network programmability. Despite the promising architecture of SDNs, security was not considered as part of the initial design. Moreover, security concerns are potentially augmented considering the logical centralization of network intelligence. Furthermore, the security and dependability of the SDN has largely been a neglected topic and remains an open issue. The paper presents a broad overview of the security implications of each SDN layer/interface. This paper contributes further by devising a contemporary layered/interface taxonomy of the reported security vulnerabilities, attacks, and challenges of SDN. We also highlight and analyze the possible threats on each layer/interface of SDN to help design secure SDNs. Moreover, the ensuing paper contributes by presenting the state-of-the-art SDNs security solutions. The categorization of solutions is followed by a critical analysis and discussion to devise a comprehensive thematic taxonomy. We advocate the production of secure and dependable SDNs by presenting potential requirements and key enablers. Finally, in an effort to anticipate secure and dependable SDNs, we present the ongoing open security issues, challenges and future research directions. <s> BIB007 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Introduction <s> Is software-defined networking (SDN) friend or foe in terms of security? Drawing on a recent Dagstuhl seminar, the authors discuss SDN's security challenges, debate strategies to monitor and protect SDN-enabled networks, and propose methods and strategies to leverage SDN's flexibility for designing new security mechanisms. <s> BIB008 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Introduction <s> Software-defined networking (SDN) emerged as an attempt to introduce network innovations faster, and to radically simplify and automate the management of large networks. SDN traditionally leverages OpenFlow as device-level abstraction. Since OpenFlow permits the programmer to “just” abstract a static flow-table, any stateful control and processing intelligence is necessarily delegated to the network controller. Motivated by the latency and signaling overhead that comes along with such a two-tiered SDN programming model, in the last couple of years several works have proposed innovative switch-level (data plane) programming abstractions capable to deploy some smartness directly inside the network switches, e.g., in the form of localized stateful flow processing. Furthermore, the possible inclusion of states and state maintenance primitives inside the switches is currently being debated in the OpenFlow standardization community itself. In this paper, after having provided the reader with a background on such emerging stateful SDN data plane proposals, we focus our attention on the security implications that data plane programmability brings about. Also via the identification of potential attack scenarios, we specifically highlight possible vulnerabilities specific to stateful in-switch processing (including denial of service and saturation attacks), which we believe should be carefully taken into consideration in the ongoing design of current and future proposals for stateful SDN data planes. <s> BIB009 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Introduction <s> Software defined networking implements the network control plane in an external entity, rather than in each individual device as in conventional networks. This architectural difference implies a different design for control functions necessary for essential network properties, e.g., loop prevention and link redundancy. We explore how such differences redefine the security weaknesses in the SDN control plane and provide a framework for comparative analysis which focuses on essential network properties required by typical production networks. This enables analysis of how these properties are delivered by the control planes of SDN and conventional networks, and to compare security risks and mitigations. Despite the architectural difference, we find similar, but not identical, exposures in control plane security if both network paradigms provide the same network properties and are analyzed under the same threat model. However, defenses vary; SDN cannot depend on edge based filtering to protect its control plane, while this is arguably the primary defense in conventional networks. Our concrete security analysis suggests that a distributed SDN architecture that supports fault tolerance and consistency checks is important for SDN control plane security. Our analysis methodology may be of independent interest for future security analysis of SDN and conventional networks. <s> BIB010

The growing demand for the efficient management and reconfiguration of large-scale communication services in computer networks, has leveraged the adoption of the network programmability paradigm of the Software Defined Email addresses: [email protected] (Tiago V. Ortiz), [email protected] (Bruno Y. L. Kimura), [email protected] (Jó Ueyama), [email protected] (Valério Rosset ) Corresponding author. Networks (SDNs) . The flexibility provided by network programmability allows several solutions to be deployed for complex tasks and thus reduces the need to replace the hardware/firmware of network devices. Among the many advantages of using SDNs, is that it enables autonomic services to be installed which are aimed at the optimization of network metrics, network virtualization, migration, mobility and energy conservation BIB004 . Unlike the case of traditional network architecture, in a SDN, the switches comprise the data plane and have two key features. First, the switches interact with a logical network control entity, known as the SDN controller, to obtain forwarding instructions. Second, they keep a flow table in which entries with instructions are stored and evaluated so that they can carry out the forwarding of packets . Generally, the function of the SDN controller is operated by non-standard software that provides the necessary interfaces for network services and management. In SDNs, the interaction between switches and controllers generally occurs through an application programming interface (API). The OpenFlow (OF) BIB001 protocol is recognized as the standard API for SDN and it is recommended by the guidelines of the Open Network Foundation (ONF) 2 . Hence, the applications hosted by the controller can implement policies, routing algorithms and other management services (e.g. load balancing, virtualization, support for mobility etc.) to determine a set of entries in the flow tables of the switches. Thus, the controller and applications act as the "brain" of the network by forming the logical network control. Security in SDN is a major concern that has recently attracted the attention of the scientific community BIB006 BIB010 . This can be explained by the fact that the programmability and centralized management of SDNs may either assist in the implementation of security services or lead to the emergence of new security threats which may compromise data and the network operations BIB005 . Comparatively, SDNs are more vulnerable to attacks than legacy networks BIB006 . It is noteworthy that the OF specifications stipulate that communication between the controller(s) and switch(es) must occur through a secure channel, that is, it must be implemented by secure transport services (e.g. TLS). As well as this, owing to the fact that many interfaces/operations in SDNs are not standardized, there is no guarantee that any single mechanism can ensure proper security. For this reason, the scientific community has dedicated a great deal of effort to investigating potential security risks in SDNs BIB008 . On the one hand, one may observe in the related literature the existence of several papers concerned with i) carrying out reviews of security in SDNs, ii) defining concepts in theoretical terms and iii) discussing possible security vulnerabilities, threats and countermeasures BIB002 BIB005 BIB009 . On the other hand, an attempt has been made to conduct experimental security analysis with focus on the detection of vulnerabilities in different planes of the SDN architecture. In addition, another important factor that should be taken into account is the set of security requirements applicable to each SDN plane. In specific terms, in the control plane context, describe the security requirements for SDN controllers, that are derived from an analysis and classification of threats defined by the ONF in . However, as there is a great heterogeneity of methods designed for experimental security evaluation in the literature, it is necessary to understand which of them can be applied to assess the fulfillment of the recommended security requirements by the controller software implementations. In view of this, we present in this paper the results of an extensive review of the most recent literature dedicated to the experimental security analysis of controller software in SDNs. Our research study complements the findings of current surveys that focus on general aspects of SDN security, e.g. BIB006 BIB005 BIB007 BIB003 , by providing a broader discussion of the techniques employed in the literature for the experimental security analysis of controller software. In addition, we introduce a taxonomy of the approaches found in the literature review and conduct a comparative analysis of them considering the security requirements defined by the ONF and the STRIDE security threat categorization model . The remainder of this paper is structured as follows. In Section 2 we describe the theoretical concepts of SDNs with focus on its architecture, main components and security threats. In Section 3, we carry out an extensive literature review with regard to the security analysis of controllers and other closely-related components of SDNs. A discussion about the data collected during the literature review is presented in Section 4. Following this, we introduce a taxonomy and a discussion of the main methods employed for the experimental security evaluation of SDN controllers in Section 5 and Section 6, respectively. Finally, we summarize the conclusions of the paper in Section 7.

Experimental Security Analysis of Controller Software in SDNs: A Review <s> SDN: Architecture and Security Threats <s> A flexible and programmable forwarding plane is essential to maximize the value of Software-Defined Networks (SDN). In this paper, we propose Protocol-Oblivious Forwarding (POF) as a key enabler for highly flexible and programmable SDN. Our goal is to remove any dependency on protocol-specific configurations on the forwarding elements and enhance the data-path with new stateful instructions to support genuine software defined networking behavior. A generic flow instruction set (FIS) is defined to fulfill this purpose. POF helps to lower network cost by using commodity forwarding elements and to create new value by enabling numerous innovative network services. We built both hardware-based and open source software-based prototypes to demonstrate the feasibility and advantages of POF. We report the preliminary evaluation results and the insights we learnt from the experiments. POF is future-proof and expressive. We believe it represents a promising direction to evolve the OpenFlow protocol and the future SDN forwarding elements. <s> BIB001 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> SDN: Architecture and Security Threats <s> The idea of programmable networks has recently re-gained considerable momentum due to the emergence of the Software-Defined Networking (SDN) paradigm. SDN, often referred to as a "radical new idea in networking", promises to dramatically simplify network management and enable innovation through network programmability. This paper surveys the state-of-the-art in programmable networks with an emphasis on SDN. We provide a historic perspective of programmable networks from early ideas to recent developments. Then we present the SDN architecture and the OpenFlow standard in particular, discuss current alternatives for implementation and testing of SDN-based protocols and services, examine current and future SDN applications, and explore promising research directions based on the SDN paradigm. <s> BIB002 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> SDN: Architecture and Security Threats <s> Currently many aspects of the classical architecture of the Internet are etched in stone - a so called ossification of the Internet - which has led to major obstacles in IPv6 deployment and difficulty in using IP multicast services. Yet, there exist many reasons to extend the Internet, e.g., for improving intra-domain and inter-domain routing for high availability of the network, providing end-to-end connectivity for users, and allowing dynamic QoS management of network resources for new applications, such as data center, cloud computing, and network virtualization. To address these requirements, the next-generation architecture for the Future Internet has introduced the concept of Software-Defined Networking (SDN). At the core of this emerging paradigm is the separation and centralization of the control plane from the forwarding elements in the network as opposed to the distributed control plane of existing networks. This decoupling allows deployment of control plane software components (e.g., OpenFlow controller) on computer platforms that are much more powerful than traditional network equipment (e.g., switches/routers) while protecting the data and intellectual property of the vendors of such equipment.A critical understanding of this emerging paradigm is necessary to address the multiple challenges in realizing the Future Internet and to resolve the ossification problem of the existing Internet. To address these requirements, this paper surveys existing technologies and the wide range of recent and state-of-the-art projects on SDN followed by an in-depth discussion of the major challenges in this area. <s> BIB003 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> SDN: Architecture and Security Threats <s> Software defined networking (SDN) decouples the network control and data planes. The network intelligence and state are logically centralized and the underlying network infrastructure is abstracted from applications. SDN enhances network security by means of global visibility of the network state where a conflict can be easily resolved from the logically centralized control plane. Hence, the SDN architecture empowers networks to actively monitor traffic and diagnose threats to facilitates network forensics, security policy alteration, and security service insertion. The separation of the control and data planes, however, opens security challenges, such as man-in-the middle attacks, denial of service (DoS) attacks, and saturation attacks. In this paper, we analyze security threats to application, control, and data planes of SDN. The security platforms that secure each of the planes are described followed by various security approaches for network-wide security in SDN. SDN security is analyzed according to security dimensions of the ITU-T recommendation, as well as, by the costs of security solutions. In a nutshell, this paper highlights the present and future security challenges in SDN and future directions for secure SDN. <s> BIB004 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> SDN: Architecture and Security Threats <s> In recent years, Software-Defined Networking (SDN) has been a focus of research. As a promising network architecture, SDN will possibly replace traditional networking, as it brings promising opportunities for network management in terms of simplicity, programmability, and elasticity. While many efforts are currently being made to standardize this emerging paradigm, careful attention needs to be also paid to security at this early design stage. This paper focuses on the security aspects of SDN. We begin by discussing characteristics and standards of SDN. On the basis of these, we discuss the security features as a whole and then analyze the security threats and countermeasures in detail from three aspects, based on which part of the SDN paradigm they target, i.e., the data forwarding layer, the control layer and the application layer. Countermeasure techniques that could be used to prevent, mitigate, or recover from some of such attacks are also described, while the threats encountered when developing these defensive mechanisms are highlighted. <s> BIB005

The purpose of the paradigm adopted by the SDNs is to decouple logical network control from packet forward by dividing the networking tasks into three planes: the data plane, the control plane and the application/management plane . Figure 1 shows an overview of the SDN architecture with the three planes included. In the SDN architecture, the interaction between the network components in each plane occurs through specific interfaces. In this case, each interface is classified as belonging to one of the four extremities of the architecture: northbound, southbound, westbound and eastbound BIB002 . The data plane essentially consists of network switches that forward frames, belonging to each flow, to other devices by means of a set of locally-installed flow rules. The SDN controller is able to configure the forward task by installing the flow rules into the switches proactively or reactively. In the reactive forwarding, on receiving the first frame of a flow, a switch usually requests the SDN controller to lay down the rules regarding that individual flow, and to keep those rules active until the end of each stream. In the proactive forwarding, the controller sends a set of rules to the switches that are defined a priori for each flow. In the absence of a rule for a specific received flow, the switch may query the controller to obtain the new rule or forward the flow in the traditional way in accordance with a prior configuration of the device. In the control plane, the controller acts as a network operating system (NOS) and provides the basic abstractions and functionalities needed for the implementation of the forward policies defined by the management entity of an Autonomous System (AS). Controllers can communicate with each other to ensure distributed network control. Alternatively, a controller can interact either with other controllers belonging to a different AS or with legacy devices by means of the westbound and eastbound interfaces, respectively BIB003 . In addition, a controller can interact with the data plane through southbound protocols. In this particular situation, although there exist several APIs and protocols designed for SDNs, (such as the Forwarding and Control Element Separation (ForCES) , the Open vSwitch Database Management Protocol (OVSDB) and the Protocol-oblivious forwarding (POF) BIB001 ), the OpenFlow protocol emerged as a de facto standard owing to its wide acceptance by practitioners and manufacturers . The application/management plane allows the network management entity to monitor and execute control operations on the SDN, usually by interacting with the controller(s) via northbound APIs/protocols, (e.g. RESTFUL APIs or programming languages ). It should be mentioned that, although the theoretical SDN architecture suggests that the application and control planes software are completely independent, this premise is rejected by the current state of SDN controller software implementations. Generally, both the control and application are a part of the same software and share computational resources. In light of this, the northbound interface is often used by the applications for communication with external entities instead of the controller. Hence , as applications may operate as plug-ins or particular modules tied to the controller software, it is not unusual to experience collateral damage to the controller software caused by the malfunction of some application. Security in SDNs is an important research field because the programmability and centralized management of SDNs may either assist in the implementation of security services or lead to the emergence of new security threats which , in turn, can compromise data and the network operation BIB004 . Comparatively, the SDNs are more vulnerable to attacks than legacy networks BIB005 . In this sense one may enumerate at least seven possible points of attack in SDNs (as shown in Figure 1 ), which are as follows: the SDN switch, the data communication between switches, the SDN controller, the southbound communication between controller(s) and the switch(es), the east/westbound communication between controllers or between application/management software, the northbound communication between the controller and the application/management software and the management/application software. It should be noted that the OpenFlow specifications state that communication between the SDN controller(s) and the switch(es) must occur through a secure channel (e.g. by using TLS). Despite this, largely owing to a lack of standardization, there are no guarantees that the current mechanisms are suitable and can meet all the security requirements of SDNs. Keeping these points in mind, in the next section, we outline the state-ofthe-art with regard to the experimental security analysis of controller software in SDNs.

Experimental Security Analysis of Controller Software in SDNs: A Review <s> Literature Review <s> This whitepaper proposes OpenFlow: a way for researchers to run experimental protocols in the networks they use every day. OpenFlow is based on an Ethernet switch, with an internal flow-table, and a standardized interface to add and remove flow entries. Our goal is to encourage networking vendors to add OpenFlow to their switch products for deployment in college campus backbones and wiring closets. We believe that OpenFlow is a pragmatic compromise: on one hand, it allows researchers to run experiments on heterogeneous switches in a uniform way at line-rate and with high port-density; while on the other hand, vendors do not need to expose the internal workings of their switches. In addition to allowing researchers to evaluate their ideas in real-world traffic settings, OpenFlow could serve as a useful campus component in proposed large-scale testbeds like GENI. Two buildings at Stanford University will soon run OpenFlow networks, using commercial Ethernet switches and routers. We will work to encourage deployment at other schools; and We encourage you to consider deploying OpenFlow in your university network too <s> BIB001 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Literature Review <s> This document specifies the Path Computation Element (PCE) ::: Communication Protocol (PCEP) for communications between a Path ::: Computation Client (PCC) and a PCE, or between two PCEs. Such ::: interactions include path computation requests and path computation ::: replies as well as notifications of specific states related to the use ::: of a PCE in the context of Multiprotocol Label Switching (MPLS) and ::: Generalized MPLS (GMPLS) Traffic Engineering. PCEP is designed to be ::: flexible and extensible so as to easily allow for the addition of ::: further messages and objects, should further requirements be expressed ::: in the future. [STANDARDS-TRACK] <s> BIB002 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Literature Review <s> The Network Configuration Protocol (NETCONF) defined in this document ::: provides mechanisms to install, manipulate, and delete the ::: configuration of network devices. It uses an Extensible Markup ::: Language (XML)-based data encoding for the configuration data as well ::: as the protocol messages. The NETCONF protocol operations are realized ::: as remote procedure calls (RPCs). This document obsoletes RFC 4741. ::: [STANDARDS-TRACK] <s> BIB003 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Literature Review <s> P4 is a high-level language for programming protocol-independent packet processors. P4 works in conjunction with SDN control protocols like OpenFlow. In its current form, OpenFlow explicitly specifies protocol headers on which it operates. This set has grown from 12 to 41 fields in a few years, increasing the complexity of the specification while still not providing the flexibility to add new headers. In this paper we propose P4 as a strawman proposal for how OpenFlow should evolve in the future. We have three goals: (1) Reconfigurability in the field: Programmers should be able to change the way switches process packets once they are deployed. (2) Protocol independence: Switches should not be tied to any specific network protocols. (3) Target independence: Programmers should be able to describe packet-processing functionality independently of the specifics of the underlying hardware. As an example, we describe how to use P4 to configure a switch to add a new hierarchical label. <s> BIB004 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Literature Review <s> The key challenges facing network architecture today are the ability to change rapidly with business needs and to control complexity. The Interface to the Routing System is one form of software-defined networks designed to address specific problems at the Internet scale. <s> BIB005

For this review, we carried out a rigorous and extensive verification of the literature by checking the bibliographic repositories available on the Internet. We selected a set of papers considering only the most relevant studies fully or partially dedicated to the experimental security analysis of the controller software in SDNs. More specifically, the literature review mainly includes studies related to the following controllers: OpenDayLight, Floodlight, Beacon, Ryu, POX, ONOS, NOX, Maestro and their variants. Whilst there exist multiple solutions for south-bound SDN interfaces, e.g., PCEP BIB002 , NETCONF BIB003 , P4 BIB004 , ForCES , I2RS BIB005 , the OpenFlow BIB001 is by far the most widespread existing solution. Thus, by examining the most recent experimental analyses of security issues with regard to open SDN controllers in the literature, we necessarily report studies related to innate OpenFlow-based SDNs. For further detail about the security analysis of other south-bound protocols in SDNs, the reader should refer to . Among the set of analyzed papers presenting practical and theoretical results, we selected only the experimental studies to be described in the following subsections. For ease of understanding, the selected papers were grouped according to the main focus of each experiment carried out, as depicted in Figure 2 , in: data plane; application and management plane; control plane and regardless the SDN plane. After describing the selected experimental studies in this section, we provide a comprehensive analysis of their security aspects in Section 4 and Section 5.

Experimental Security Analysis of Controller Software in SDNs: A Review <s> Security Analysis in the Data Plane <s> SDN is a concept of programmable networking. A network manager can process each network flow through software programs. There is a hypothesis that all switches are trusted and follow programmed commands to handle packets. That is, once a switch is compromised by an attacker and does not follow the order of the network manager, this will bring a huge network disaster. In this paper, we define some attack models through compromised switches and design a detection mechanism to find these compromised devices. We evaluate our mechanism and discuss some future works. <s> BIB001 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Security Analysis in the Data Plane <s> Software-Defined Networking (SDN) allows scalable and flexible network management without requiring expensive hardware changes. However, this technology is relatively new and creates novel security risks. In this paper, we investigate the vulnerability of link discovery service in SDN controller, which is a critical service provided by the control layer for the proper functioning of applications and network services. We also discuss the potential attacks on link discovery service, which are verified via experiments. <s> BIB002

In this section we describe the experimental security analysis with regard to the data plane, presented in , BIB001 and BIB002 , respectively. The authors of analyzed the effects of overhead on the routing tables of the SDN switches. As an experiment, the authors implemented a scenario in which an attack was carried out by continuously sending several flows of forged raw packets. In this case, only a small portion of the packet header was changed between streams, so that an individual entry could be inserted into the switch for each forged packet until the storage capacity was exhausted. According to the authors of , when the resources of the switch are exhausted, the controller is unable to install new entries for legitimate flows. To mitigate this type of attack, they recommend that the controller should either wait until a flow entry expires before inserting a new entry of a legitimate flow or directly forward the legitimate flow through to another switch. The authors also conducted tests to measure the delay, throughput, and packet drop rate when the flow table is exhausted. The measurements were performed on the OpenDayLight (ODL) controller that was configured with the two mitigation strategies. The tests considered legitimate flows with only a subset of entries already installed in the switch and forged flows with no related entries in the switch tables. As a result, the evaluation detected a performance degradation in the controller. In addition, the tests also revealed a degradation in the throughput of the legitimate flows when there were no entries installed in the switch; this was mainly due to the extra work carried out by the controller to forward the packets to other switches. The authors of BIB001 introduced a prevention scheme against attacks originating in compromised OpenFlow switches. A system scenario is defined in which a compromised switch is able to launch three types of active attacks, namely, incorrect forwarding (deviation or duplication of a flow), packet manipulating and a malicious weight adjustment of the compromised switch group tables. The authors created two algorithms to detect these attack vectors. The purpose of the first algorithm, called Forwarding Detection, is to verify if the flows are being correctly routed by the switches. To do this , the algorithm uses an artificial packet to check the flow entries in the switch tables and tracks the path traveled by the flow to determine the correctness of the path. The second algorithm, called Weighting Detection, is used to check if the switch forwards packets at an expected rate. The algorithm generates and sends a set of artificial packets to the evaluated switches and analyses the amount of received packets. If the number of received packets is close to the expected value, the flow is considered to be satisfactory; otherwise the flow is compromised. Recently, in BIB002 the authors evaluated the link discovery service (LDS) vulnerability, which is an essential service of the control plane that is needed for the proper operation of SDN services and applications. The authors provided details of how an attacker can "poison" the perception of the SDN controller about the network topology, by introducing false links. They also explained that such links can be created through the use of false Link Layer Discovery Protocol (LLDP) messages, sent by one or more compromised hosts. As a result, the authors were able to simulate a scenario of a realistic link falsification attack on the FloodLight controller implemented on the Mininet emulator.

Experimental Security Analysis of Controller Software in SDNs: A Review <s> Security Analysis in the Control Plane <s> Software-defined networks facilitate rapid and open innovation at the network control layer by providing a programmable network infrastructure for computing flow policies on demand. However, the dynamism of programmable networks also introduces new security challenges that demand innovative solutions. A critical challenge is efficient detection and reconciliation of potentially conflicting flow rules imposed by dynamic OpenFlow (OF) applications. To that end, we introduce FortNOX, a software extension that provides role-based authorization and security constraint enforcement for the NOX OpenFlow controller. FortNOX enables NOX to check flow rule contradictions in real time, and implements a novel analysis algorithm that is robust even in cases where an adversarial OF application attempts to strategically insert flow rules that would otherwise circumvent flow rules imposed by OF security applications. We demonstrate the utility of FortNOX through a prototype implementation and use it to examine performance and efficiency aspects of the proposed framework. <s> BIB001 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Security Analysis in the Control Plane <s> This paper presents an independent comprehensive analysis of the efficiency indexes of popular open source SDN/OpenFlow controllers (NOX, POX, Beacon, Floodlight, MuL, Maestro, Ryu). The analysed indexes include performance, scalability, reliability, and security. For testing purposes we developed the new framework called hcprobe. The test bed and the methodology we used are discussed in detail so that everyone could reproduce our experiments. The result of the evaluation show that modern SDN/OpenFlow controllers are not ready to be used in production and have to be improved in order to increase all above mentioned characteristics. <s> BIB002 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Security Analysis in the Control Plane <s> Software-Defined Networking proposes a new paradigm by separating the control plane from the data plane. The control plane's main responsibility is deployed in a centralized component called SDN controller. Therefore, the SDN based network becomes dependent on a single functional entity (although it may have multiple instances) in order to accomplish its routing and policy rules. The SDN controller is vulnerable to degradation due to a well-known phenomenon called Software Aging. To the best of our knowledge, this work is the first that assesses the impacts of software aging on current implementation of SDN controllers. We show that such a phenomenon must not be neglected in future software architecture designs and implementations. <s> BIB003 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Security Analysis in the Control Plane <s> One of the core properties of Software Defined Networking (SDN) is the ability for third parties to develop network applications. This introduces increased potential for innovation in networking from performance-enhanced to energy-efficient designs. In SDN, the application connects with the network via the SDN controller. A specific concern relating to this communication channel is whether an application can be trusted or not. For example, what information about the network state is gathered by the application? Is this information necessary for the application to execute or is it gathered for malicious intent? In this paper we present an approach to secure the northbound interface by introducing a permissions system that ensures that controller operations are available to trusted applications only. Implementation of this permissions system with our Operation Checkpoint adds negligible overhead and illustrates successful defense against unauthorized control function access attempts. <s> BIB004 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Security Analysis in the Control Plane <s> Both academic researchers and network vendors have started to promote Software-Defined Networking (SDN) as a new network paradigm, in which controller systems play a major role. A modular and extensible design allows network operators to extend the controller's functionality by so called network services. Unfortunately, in current designs such network services have unlimited access to mandatory SDN resources which enables different kinds of attacks. To retain control over network services (especially third-party ones), we adapt approved security mechanisms and propose a containment mechanism as well as a framework to ease containment configuration. For both proposals, we provide proof-of-concept implementations for an open and industry-supported reference framework and hereby aim to improve security for a wide range of SDN controllers. Finally, our proposals achieve the ability to harden a mandatory SDN component (i. e., the SDN controller) and enable proactive security even against malicious network services. <s> BIB005 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Security Analysis in the Control Plane <s> In this paper, we investigated the vulnerabilities surrounding software-defined networking (SDN). Specifically, we examined the vulnerabilities of OpenDayLight SDN controller. Among all the possible attack vectors, a man-in-the-middle attack using ARP poisoning was successfully launched to intercept the traffic between a client and the OpenDayLight controller. Details of the experiment method, procedures and results were described in this manuscript. To the best of our knowledge, this is the first successful practical attempt to penetrate an SDN controller and be able to capture login credentials of the controller. The significance of this attack should not be taken lightly; once the SDN controller is under the control of the adversary, there will be no security at all for the entire network governed by this controller. <s> BIB006

Here we describe the approaches employed for assessing security in the control plane and comparing the communication interfaces with the other planes in SDNs. The authors of BIB003 investigate the impact of software aging (SA) on SDN controllers. The concept of SA was used to analyze the performance and security of a system by observing the operating conditions of a software over an extended period of time. The authors analyzed the impact of the SA on the FloodLight and Beacon controllers. In the experiments, the controllers remained in operation for 36 hours, with alternating periods of high workload and idleness lasting for twenty and five minutes, respectively. The authors used the cbench tool to generate the workload (packet-in flows) for 16 Openflow switches at a rate of 100.000 flows/s. As a result, it was found that FloodLight used all the available resident memory (4.2GB) of the Java Virtual Machine (JVM) as well as all the virtual memory, including that stored on disk, making a total of 8GB of memory consumption. The memory management of Beacon was better since it used approximately 500MB of resident memory and 5GB of virtual memory. The study carried out by BIB004 is concerned with unauthorized access to SDN controllers including threats originating from applications. The authors propose a set of Authorization and Access Control rules as well as a reference monitor named OperationCheckpoint, which is coupled to the FloodLight controller. The improvements include the following: a) the definition of a set of rules for the OpenFlow commands of a given application, b) the creation of unique identifiers (IDs) for applications, c) Permission Management and d) persistent records of actions and access to the controller. The authors validated the scheme by evaluating the OperationCheckpoint in an environment emulated with Mininet. The experiments were divided into two scenarios: one considering access to the REST API and another considering access to the internal Java methods implemented in the controller. The proposed OperationCheckpoint prevented access to calls for the REST API methods, but allowed calls made through Java methods due to the internality of Java. Finally, the authors measured the latency that was introduced by using the OperationCheckpoint. There was an average increase of approximately 367.125µs in the latency of the operations performed by the controller. Similarly to BIB004 , the authors of BIB005 designed a configurable mechanism to restrict access to system-level operations through the application of security policies. The purpose of the system is to maintain control over third-party applications installed in the SDN controller and thus prevent attacks resulting from malicious actions. In the mechanism, each network service that can carry out sensitive operations, is held in a sandbox. Two modes of operation were defined: detection and protection. In the detection mode, the operations are monitored and the security breaches are only recorded in the log file. In the protection mode, sandboxes can restrict the execution of operations by complying with the security rules defined by the network administrator. The mechanism was implemented and validated in the OpenDaylight controller within the Mininet environment. The authors validated the mechanism by conducting effectiveness and performance tests. The purpose of the effectiveness test was to determine whether the security rules were being enforced for the strict execution of operations from two different applications. It was found that the mechanism was able to contain the malicious operations satisfactorily in the tests. As a result of the performance test, it was verified a reduction of less than 5% in the effective transmission rate of the network due the control overload introduced by the proposed mechanism. The authors of BIB006 investigated the vulnerabilities in the OpenDayLight controller with regard to Man-in-the-Middle (MitM) attacks. The study addresses two variants of threat vectors, namely, threats in the communication of the control plane and threats in the controller. The authors presented a MitM attack, which intercepted messages between the SDN controller and a remote client, with the aim of capturing the credentials of the controller. When carrying out the attack, the authors assumed a scenario where the adversary is located on the same LAN as the SDN controller and can capture packets. For this reason, they executed the MitM attack on the communication of a remote terminal and the OpenDayLight controller via the web DLUX interface of the controller. In the experiment, the adversary captured HTTP (non-SSL) messages exchanged between the web DLUX interface and the remote terminal. As a result, the credentials of the controller were captured in plain text. BIB002 conducted an experimental performance analysis of the respective open source controllers: NOX, POX, Beacon, FloodLight, MuL, Maestro and Ryu. The selected controllers were compared in terms of performance, scalability, reliability and security metrics. The flow rate and latency metrics were used to measure performance and scalability and covered scenarios in which the number of hosts ranged from 1 to 256 and the number of switches from 10 3 to 10 7 . In the case of the reliability analysis, the authors measured the number of connection failures and packet losses that occurred during an operational period of 24 hours by examining workload values between 2, 000 and 18, 000 requests per second. Finally, when investigating security issues, five types of malformed OpenFlow packets were created to observe the behavior of the controller on receiving these packets. As result of the experiments, it was noted that the Beacon controller had a higher flow rate while the POX, NOX and Ryu had a very low flow rate. In addition, Mul and Beacon had the lowest latencies. With regard to reliability, Mul and Maestro presented higher packet drop rate than the other controllers. Finally, despite having a poor performance, RYU achieved the best security classification level. The authors of BIB001 designed a new extension for the NOX controller, called FortNOX, which implements a role-based authorization strategy, security restrictions and conflicts resolution between flow rules. FortXOX enables NOX to check for real-time flow rule conflict. FortNOX employs a rule-based representation method, called alias reduced rules (ARRs), and detects conflicts by comparing the set of ARRs and the new installed rules. The authors validated the scheme by estimating the computational time aggregated by the conflict analysis and resolution employed in the FortNOX. As a result, the FortNOX conflict analysis showed , on average, an increase of 7ms in the processing time for each new flow rule request. The authors of introduced a security extension to the control plane to provide security management and arbitration of conflicts which originated from the insertion of multiple flow rules by distinct applications. The proposed extension layer, called the Security Enforcement Kernel (SEK), was implemented on top of the FloodLight controller. The SEK conducts a Rule-based Conflict Analysis (RCA) which acts as a conflict resolution mechanism that mediates all the requests for installing new flow rules. The SEK has the following components: an code authentication module; conflict detection and resolution; a state manager; callback tracking; a permitting mediator; a flow policy synchronizer; the separation of processes and a security audit service. [39] present several attacks on controllers that violate the network topology and the forwarding strategy in the data plane of SDNs. In addition, it demonstrated the feasibility of launching attacks on four controllers: Maestro, POX, OpenDayLight and FloodLight. For this reason, the authors proposed a new controller, -SPHINX. This controller is capable of detecting attacks by dynamically learning new network behavior and giving an alert when it detects a suspicious activity. In the experiments, the authors evaluate the controllers with regard to the following attacks: ARP poisoning, Fake topology, Controller DoS, Network DoS, TCAM exhaustion and Switch blackhole. As a result, they observed the lack of resilience of OpenDayLight, FloodLight and POX to TCAM exhaustion, DoS and Fake topology attacks, respectively. They report the resilience of SPHINX to all the attacks carried out. Additionally, with respect to performance, the results show that SPHINX can verify 1, 000 policies at each network update in ≈ 245µs, by generating an increase of ≈ 6% and ≈ 14.5% in the processing and memory usage, respectively.

Experimental Security Analysis of Controller Software in SDNs: A Review <s> Network Fingerprinting Threats <s> In this paper, for the first time we show a new attack to fin- gerprint SDN networks and further launch efficient resource consumption attacks. This attack demonstrates that SDN brings new security issues that may not be ignored. We provide the first feasibility study of such attack and hope to stimulate further studies in SDN security research. <s> BIB001 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Network Fingerprinting Threats <s> Software-defined networking (SDN) eases network management by centralizing the control plane and separating it from the data plane. The separation of planes in SDN, however, introduces new vulnerabilities in SDN networks, since the difference in processing packets at each plane allows an adversary to fingerprint the network’s packet-forwarding logic. In this paper, we study the feasibility of fingerprinting the controller-switch interactions by a remote adversary, whose aim is to acquire knowledge about specific flow rules that are installed at the switches. This knowledge empowers the adversary with a better understanding of the network’s packet-forwarding logic and exposes the network to a number of threats. In this paper, we collect measurements from hosts located across the globe using a realistic SDN network comprising of OpenFlow hardware and software switches. We show that, by leveraging information from the RTT and packet-pair dispersion of the exchanged packets, fingerprinting attacks on SDN networks succeed with overwhelming probability. We additionally show that these attacks are not restricted to active adversaries, but can also be mounted by passive adversaries that only monitor traffic exchanged with the SDN network. Finally, we discuss the implications of these attacks on the security of SDN networks, and we present and evaluate an efficient countermeasure to strengthen SDN networks against fingerprinting. Our results demonstrate the effectiveness of our countermeasure in deterring fingerprinting attacks on SDN networks. <s> BIB002

The authors of , BIB002 and BIB001 address the problem of the fingerprinting of controller-switch interactions in SDNs. By fingerprinting the SDN, an adversary can estimate the packet-forwarding logic, map the network topology and identify controllers, which are all factors that may expose the network to threats. In and BIB002 , the authors investigated the ability of a remote adversary to determine (passively or actively) if any flow rule installation has been triggered by a given packet. The rule of recognition attack is carried out by releasing new streams with a cross-traffic generator. The results are obtained by means of the statistical analysis of the Round-Trip Time (RTT) and packet-pair dispersion of these flows. The dispersion between packet pairs refers to the time interval between the complete transmission of two packets sent by a client on a particular link . When conducting the statistical analysis, the authors consider two scenarios: in the first scenario, the packet does not trigger a new installation of rules, while in the second scenario the packet does. In the experiments, it was possible to determine that a packet is able to trigger the installation of a new flow with an accuracy of 98.54%. In BIB001 , authors devised a network scanning tool called SDN scanner for remotely fingerprinting a SDN network. The SDN scanner sends fake packets to a particular target network and repeatedly estimates the response time of each packet. In this way, the tool can determine whether flows are new or already exist. Subsequently, the scanner statistically tests the samples from the two sets to determine whether a given network is SDN or not. This can be carried out by means of statistical tests, such as t-test . As a result of the experiments, the tool achieved a fingerprinting rate of 85.7%.

Experimental Security Analysis of Controller Software in SDNs: A Review <s> Network Topology Threats <s> Software Defined Networking (SDN) has been proposed as a drastic shift in the networking paradigm, by decoupling network control from the data plane and making the switching infrastructure truly programmable. The key enabler of SDN, OpenFlow, has seen widespread deployment on production networks and its adoption is constantly increasing. Although openness and programmability are primary features of OpenFlow, security is of core importance for real-world deployment. In this work, we perform a security analysis of OpenFlow using STRIDE and attack tree modeling methods, and we evaluate our approach on an emulated network testbed. The evaluation assumes an attacker model with access to the network data plane. Finally, we propose appropriate counter-measures that can potentially mitigate the security issues associated with OpenFlow networks. Our analysis and evaluation approach are not exhaustive, but are intended to be adaptable and extensible to new versions and deployment contexts of OpenFlow. <s> BIB001 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Network Topology Threats <s> Topology Discovery is an essential service in Software Defined Networks (SDN). Most SDN controllers use a de-facto standard topology discovery mechanism based on Open-Flow to identify active links in the network. This paper discusses the security, or rather lack thereof, of the current SDN topology discovery mechanism, and its vulnerability to link spoofing attacks. The feasibility and impact of the attacks are verified and demonstrated via experiments. The paper presents and evaluates a countermeasure based on HMAC authentication. <s> BIB002 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Network Topology Threats <s> Abstract Logically centralized nature of the controller in Software-defined Networking (SDN) makes it vulnerable to various adversarial attacks. These attacks have capabilities to degrade the performance of the managed network or bring it down in the worst case. For this reason, a large-scale deployment of SDN needs evaluation of the impact of adversarial attacks on network services. In this work, we implement various attacks in SDN and analyze their impact on the performance of web-based services running over it. Prior to that, we briefly discuss different kinds of vulnerabilities and threats in SDN. We consider the connection set-up latency and loss, for web-client requests, as metrics for the evaluation. We observe a significant degradation in the performances of web services, in terms of response time and availability, in the presence of implemented attacks. <s> BIB003

The authors of BIB001 evaluated the security in SDNs with a special focus on the OpenFlow protocol. In their study, the OpenFlow is analyzed on the basis of the STRIDE security threat categorization model. When conducting the analysis, a data flow diagram (DFD ) is used to model the OpenFlow so that it can be combined with the vulnerabilities of the STRIDE threat categorization model, and thus result in several Attack Trees. After the trees have been generated, practical tests are carried out to validate the approach, such as: Denial of Service (DoS) in the switch flow table, Denial of Service (DoS) in the switch input buffer and fingerprinting the network to determine if a rule exists on a switch and if the activity of a particular client can be tracked. [44] assessed the security of the Open Flow Discovery Protocol (OFDP), a variation of the Link Layer Discovery Protocol (LLDP). The OFDP is part of the OpenFlow software distribution. However, OFDP does not support authentication, integrity or confidentiality. Thus , it is susceptible to network topology poisoning attacks through spoofed links. The authors of BIB002 conducted experiments in which it was possible to create unidirectional and bidirectional false links by using one and two infected hosts, respectively. In another scenario consisting of five switches and five client hosts, the network connectivity is reduced by 30% through the creation of one false link. As an improvement, the authors suggest the use of a Keyed-Hash Message Authentication Code (HMAC) to ensure authenticity and integrity in the network. The authors report that the use of OFDP with HMAC increases the network processing load by approximately 8%. The authors of examine attack vectors for SDNs based on the exploitation of false links. The attacks are intended to poison the topology of the network to mislead the core services of the controllers and applications. The authors found failures in the Host Tracking Service and Link Discovery Service systems in the SDN controllers. This led to a pair of poisoning attack groups in the network, namely Host Location Hijacking and Link Fabrication attacks. The Host Location Hijacking attack group encompasses Exploitation in Host Tracking Service (HTS) attacks, in which an adversary impersonates a legitimate host to capture the network traffic destined for the legitimate host, and the Web Clients Harvesting, where the attacker impersonates a legitimate web server. The Link Fabrication attack includes five variations: Exploitation in Link Discovery Service, Fake LLDP Injection, LLDP Relay, Denial of Service Attack, and MitM. The proposed attacks were successfully carried out on the FloodLight, OpenDayLight, Beacon and POX controllers. As an improvement, the authors added a new security extension to the Floodlight controller, called TopoGuard, which provides automatic and real-time detection of Network Topology Poisoning Attacks. The study BIB003 analyses the impact of attacks to SDNs through the latency and packet loss observed in web services traffic. Without pointing out possible countermeasures, the authors discuss vulnerabilities and security threats, by highlighting the following : threats to SDN management, threats to the control plane and threats to the data plane. Two types of attacks were carried out. In the first attack, an adversary sends false requests as a means of keeping the controllers and switches busy, as well as causing delay and data loss. In the second attack, a malicious switch is employed to impersonate a legitimate switch, which leads to a loss of connection with the SDN controller.

Experimental Security Analysis of Controller Software in SDNs: A Review <s> Trust and Risk Assessment <s> Software-defined networking empowers network operators with more flexibility to program their networks. With SDN, network management moves from codifying functionality in terms of low-level device configurations to building software that facilitates network management and debugging. By separating the complexity of state distribution from network specification, SDN provides new ways to solve long-standing problems in networking --- routing, for instance --- while simultaneously allowing the use of security and dependability techniques, such as access control or multi-path. However, the security and dependability of the SDN itself is still an open issue. In this position paper we argue for the need to build secure and dependable SDNs by design. As a first step in this direction we describe several threat vectors that may enable the exploit of SDN vulnerabilities. We then sketch the design of a secure and dependable SDN control platform as a materialization of the concept here advocated. We hope that this paper will trigger discussions in the SDN community around these issues and serve as a catalyser to join efforts from the networking and security & dependability communities in the ultimate goal of building resilient control planes. <s> BIB001 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Trust and Risk Assessment <s> The SDN paradigm allows networks to be dynamically reconfigurable by network applications. SDN is also of particular interest for NFV which deals with the virtualization of network functions. The network programmability offered by SDN presents then various advantages but it also induces various threats regarding potential attacks on the network. For instance, there is a critical risk that a hacker takes over the network control by exploiting this SDN network programmability (e.g., using the SDN API or tampering a network application running on the SDN controller). This paper proposes then an approach to deal with this possible lack of trust in the SDN controller or in their applications. This approach consists in not relying on a single controller but on several ‘redundant’ controllers that may also run in different execution environments. The network configuration requests coming from these controllers are then compared and, if deemed sufficiently consistent and then trustable, they are actually sent to the network. This approach has been implemented in an intermediary layer (based on a network hypervisor) inserted between the network equipments and the controllers. Experimentations have been performed showing the feasibility of the approach and providing some first evaluations of its impact on the network and the services. <s> BIB002 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Trust and Risk Assessment <s> Software-defined networking (SDN) enables the exible and dynamic configuration of a network, and OpenFlow is one practical SDN implementation. Although it has been widely deployed in actual environments, it can cause fatal aws. In this paper, we consolidate the security threats to OpenFlow mentioned in previous work and introduce a new security checksheet that includes risk assessment methods. We compare the Kreutz et al. threat vectors with the SDNSecurity.org attack list to discover new threats. Our checksheet enables the security of a given OpenFlow network design to be comprehensively assessed. Furthermore, we evaluate the performance of an OpenFlow network with two attack scenarios using the checksheet and identify critical performance degradations. <s> BIB003

The study conducted by BIB002 deals with the question of trust analysis in SDN controllers and their applications. The introduced approach relies on several redundant controllers that are capable of operating in different environments. The authors introduce an intermediate SDN layer that is located between the switches and the controllers, called Trust-oriented Controller Proxy (ToCP), and based on a hypervisor. The ToCP layer collects and analyses the rule installation requests originating from distinct controllers and if they are deemed consistent and trustworthy, it installs the rules on the network switches. The authors made an evaluation of ToCP in a SDN network by taking note of the throughput, packet loss, Jitter, memory and processing overheads. More specifically, the functionalities of ToCP increased the Jitter and packet loss and reduced the throughput by 13.6% in the worst case scenario. Similarly, the memory consumption and CPU usage were also increased by 401MB and 0.4%, respectively. In BIB003 , the authors classified security threats by introducing a method of risk assessment and countermeasures for some of the security issues pointed out. They drew up a security checksheet to help network designers to determine risks and find suitable countermeasures to mitigate them. The list of attacks was combined with a prior list described in BIB001 . As an experiment, the authors first made a qualitative assessment of an SDN testbed by using the proposed checksheet. Afterwards, two Denial of Service (DoS) attacks were carried out to quantitatively measure the performance degradation of the network.

Experimental Security Analysis of Controller Software in SDNs: A Review <s> Availability threats <s> Distributed denial of service (DDoS) attacks in cloud computing environments are growing due to the essential characteristics of cloud computing. With recent advances in software-defined networking (SDN), SDN-based cloud brings us new chances to defeat DDoS attacks in cloud computing environments. Nevertheless, there is a contradictory relationship between SDN and DDoS attacks. On one hand, the capabilities of SDN, including software-based traffic analysis, centralized control, global view of the network, dynamic updating of forwarding rules, make it easier to detect and react to DDoS attacks. On the other hand, the security of SDN itself remains to be addressed, and potential DDoS vulnerabilities exist across SDN platforms. In this paper, we discuss the new trends and characteristics of DDoS attacks in cloud computing, and provide a comprehensive survey of defense mechanisms against DDoS attacks using SDN. In addition, we review the studies about launching DDoS attacks on SDN, as well as the methods against DDoS attacks in SDN. To the best of our knowledge, the contradictory relationship between SDN and DDoS attacks has not been well addressed in previous works. This work can help to understand how to make full use of SDN's advantages to defeat DDoS attacks in cloud computing environments and how to prevent SDN itself from becoming a victim of DDoS attacks, which are important for the smooth evolution of SDN-based cloud without the distraction of DDoS attacks. <s> BIB001 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Availability threats <s> Software Defined Networking (SDN) provides a simpler and more efficient approach to manage computer networks. This is achieved by decoupling the control functionality from the forwarding elements, and placing it into a logically centralised controller. This centralised architecture makes SDN susceptible to a range of Denial of Service (DoS) attacks. In this paper we investigate a range of DoS attacks against OpenFlow-based SDN, both against the control plane and data plane, and quantify the impact of the attacks via experiments. In particular, we compare the impact on three key SDN controller platforms, i.e. ONOS, Floodlight and Ryu. Our results show that with relatively limited resources, an attacker can cause a significant disruption against an SDN. <s> BIB002

It is noteworthy that a great number of papers in the literature address the availability of resources by exploring Denial of Service (DoS) attacks in SDNs. Among them, we selected one study to represent this category in this review, BIB002 . A comprehensive review of Distributed DoS attacks on SDNs can be found in BIB001 . In BIB002 , the authors carry out an experimental evaluation of the impact of DoS attacks on SDNs by looking at two types of DoS attacks against the control and data planes. In the data plane, they examine two types of attacks aimed at exhausting both the resources of the SDN controller and switches. The authors concluded that an adversary controlling a single host is able to disrupt the forwarding capability of a SDN network with relatively limited resources.

Experimental Security Analysis of Controller Software in SDNs: A Review <s> Paper <s> Software-defined networks facilitate rapid and open innovation at the network control layer by providing a programmable network infrastructure for computing flow policies on demand. However, the dynamism of programmable networks also introduces new security challenges that demand innovative solutions. A critical challenge is efficient detection and reconciliation of potentially conflicting flow rules imposed by dynamic OpenFlow (OF) applications. To that end, we introduce FortNOX, a software extension that provides role-based authorization and security constraint enforcement for the NOX OpenFlow controller. FortNOX enables NOX to check flow rule contradictions in real time, and implements a novel analysis algorithm that is robust even in cases where an adversarial OF application attempts to strategically insert flow rules that would otherwise circumvent flow rules imposed by OF security applications. We demonstrate the utility of FortNOX through a prototype implementation and use it to examine performance and efficiency aspects of the proposed framework. <s> BIB001 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Paper <s> Software Defined Networking (SDN) has been proposed as a drastic shift in the networking paradigm, by decoupling network control from the data plane and making the switching infrastructure truly programmable. The key enabler of SDN, OpenFlow, has seen widespread deployment on production networks and its adoption is constantly increasing. Although openness and programmability are primary features of OpenFlow, security is of core importance for real-world deployment. In this work, we perform a security analysis of OpenFlow using STRIDE and attack tree modeling methods, and we evaluate our approach on an emulated network testbed. The evaluation assumes an attacker model with access to the network data plane. Finally, we propose appropriate counter-measures that can potentially mitigate the security issues associated with OpenFlow networks. Our analysis and evaluation approach are not exhaustive, but are intended to be adaptable and extensible to new versions and deployment contexts of OpenFlow. <s> BIB002 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Paper <s> This paper presents an independent comprehensive analysis of the efficiency indexes of popular open source SDN/OpenFlow controllers (NOX, POX, Beacon, Floodlight, MuL, Maestro, Ryu). The analysed indexes include performance, scalability, reliability, and security. For testing purposes we developed the new framework called hcprobe. The test bed and the methodology we used are discussed in detail so that everyone could reproduce our experiments. The result of the evaluation show that modern SDN/OpenFlow controllers are not ready to be used in production and have to be improved in order to increase all above mentioned characteristics. <s> BIB003 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Paper <s> In this paper, for the first time we show a new attack to fin- gerprint SDN networks and further launch efficient resource consumption attacks. This attack demonstrates that SDN brings new security issues that may not be ignored. We provide the first feasibility study of such attack and hope to stimulate further studies in SDN security research. <s> BIB004 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Paper <s> Software-Defined Networking proposes a new paradigm by separating the control plane from the data plane. The control plane's main responsibility is deployed in a centralized component called SDN controller. Therefore, the SDN based network becomes dependent on a single functional entity (although it may have multiple instances) in order to accomplish its routing and policy rules. The SDN controller is vulnerable to degradation due to a well-known phenomenon called Software Aging. To the best of our knowledge, this work is the first that assesses the impacts of software aging on current implementation of SDN controllers. We show that such a phenomenon must not be neglected in future software architecture designs and implementations. <s> BIB005 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Paper <s> One of the core properties of Software Defined Networking (SDN) is the ability for third parties to develop network applications. This introduces increased potential for innovation in networking from performance-enhanced to energy-efficient designs. In SDN, the application connects with the network via the SDN controller. A specific concern relating to this communication channel is whether an application can be trusted or not. For example, what information about the network state is gathered by the application? Is this information necessary for the application to execute or is it gathered for malicious intent? In this paper we present an approach to secure the northbound interface by introducing a permissions system that ensures that controller operations are available to trusted applications only. Implementation of this permissions system with our Operation Checkpoint adds negligible overhead and illustrates successful defense against unauthorized control function access attempts. <s> BIB006 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Paper <s> Within the hierarchy of the Software Defined Network (SDN) network stack, the control layer operates as the critical middleware facilitator of interactions between the data plane and the network applications, which govern flow routing decisions. In the OpenFlow implementation of the SDN model, the control layer, commonly referred to as a network operating system (NOS), has been realized by a range of competing implementations that offer various performance and functionality advantages: Floodlight, POX, NOX, and ONIX. In this paper we focus on the question of control layer resilience, when rapidly developed prototype network applications go awry, or third-party network applications incorporate unexpected vulnerabilities, fatal instabilities, or even malicious logic. We demonstrate how simple and common failures in a network application may lead to loss of the control layer, and in effect, loss of network control. To address these concerns we present the ROSEMARY controller, which implements a network application containment and resilience strategy based around the notion of spawning applications independently within a micro-NOS. ROSEMARY distinguishes itself by its blend of process containment, resource utilization monitoring, and an application permission structure, all designed to prevent common failures of network applications from halting operation of the SDN Stack. We present our design and implementation of ROSEMARY, along with an extensive evaluation of its performance relative to several of the mostly well-known and widely used controllers. Rather than imposing significant performance costs, we find that with the integration of two optimization features, ROSEMARY offers a competitive performance advantage over the majority of other controllers. <s> BIB007 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Paper <s> SDN is a concept of programmable networking. A network manager can process each network flow through software programs. There is a hypothesis that all switches are trusted and follow programmed commands to handle packets. That is, once a switch is compromised by an attacker and does not follow the order of the network manager, this will bring a huge network disaster. In this paper, we define some attack models through compromised switches and design a detection mechanism to find these compromised devices. We evaluate our mechanism and discuss some future works. <s> BIB008 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Paper <s> Both academic researchers and network vendors have started to promote Software-Defined Networking (SDN) as a new network paradigm, in which controller systems play a major role. A modular and extensible design allows network operators to extend the controller's functionality by so called network services. Unfortunately, in current designs such network services have unlimited access to mandatory SDN resources which enables different kinds of attacks. To retain control over network services (especially third-party ones), we adapt approved security mechanisms and propose a containment mechanism as well as a framework to ease containment configuration. For both proposals, we provide proof-of-concept implementations for an open and industry-supported reference framework and hereby aim to improve security for a wide range of SDN controllers. Finally, our proposals achieve the ability to harden a mandatory SDN component (i. e., the SDN controller) and enable proactive security even against malicious network services. <s> BIB009 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Paper <s> A lot of new features and capabilities emerge in the network because of separation of data plane and control plane in Software Defined Network (SDN) terminology. One of them is the possibility to implement Northbound API to allow third-party application to access resources of the network. However, most of the current implementations of Northbound API do not consider the security aspect. Therefore, we design more secure scheme for it. The design consists of token authentication for application and user, who is responsible to control/use the application/network, using OAuth 2.0 protocol. <s> BIB010 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Paper <s> Topology Discovery is an essential service in Software Defined Networks (SDN). Most SDN controllers use a de-facto standard topology discovery mechanism based on Open-Flow to identify active links in the network. This paper discusses the security, or rather lack thereof, of the current SDN topology discovery mechanism, and its vulnerability to link spoofing attacks. The feasibility and impact of the attacks are verified and demonstrated via experiments. The paper presents and evaluates a countermeasure based on HMAC authentication. <s> BIB011 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Paper <s> The SDN paradigm allows networks to be dynamically reconfigurable by network applications. SDN is also of particular interest for NFV which deals with the virtualization of network functions. The network programmability offered by SDN presents then various advantages but it also induces various threats regarding potential attacks on the network. For instance, there is a critical risk that a hacker takes over the network control by exploiting this SDN network programmability (e.g., using the SDN API or tampering a network application running on the SDN controller). This paper proposes then an approach to deal with this possible lack of trust in the SDN controller or in their applications. This approach consists in not relying on a single controller but on several ‘redundant’ controllers that may also run in different execution environments. The network configuration requests coming from these controllers are then compared and, if deemed sufficiently consistent and then trustable, they are actually sent to the network. This approach has been implemented in an intermediary layer (based on a network hypervisor) inserted between the network equipments and the controllers. Experimentations have been performed showing the feasibility of the approach and providing some first evaluations of its impact on the network and the services. <s> BIB012 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Paper <s> In this paper, we investigated the vulnerabilities surrounding software-defined networking (SDN). Specifically, we examined the vulnerabilities of OpenDayLight SDN controller. Among all the possible attack vectors, a man-in-the-middle attack using ARP poisoning was successfully launched to intercept the traffic between a client and the OpenDayLight controller. Details of the experiment method, procedures and results were described in this manuscript. To the best of our knowledge, this is the first successful practical attempt to penetrate an SDN controller and be able to capture login credentials of the controller. The significance of this attack should not be taken lightly; once the SDN controller is under the control of the adversary, there will be no security at all for the entire network governed by this controller. <s> BIB013

Controller (original) Contribution OpenDayLight Floodlight v0.9 BIB005 Floodlight e Beacon BIB008 Ryu BIB002 POX BIB006 Floodlight OperationCheckpoint BIB009 FloodLight, Beacon, OpenDayLight and HP VAN SDN Application containment mechanism with sandboxes over OpenDaylight Hydrogen BIB010 FloodLight, ONOS, OpenDayLight, Ryu and POX BIB011 POX BIB012 Floodlight v0.9 BIB013 OpenDayLight Helium BIB003 NOX, POX, Beacon, Floodlight, MuL, Maestro and Ryu BIB004 NOX, Beacon and Maestro BIB007 Floodlight, OpenDaylight, POX and Beacon Rosemary Floodlight, OpenDaylight and ONOS BIB001 NOX 0.9. be grouped into three categories : simulated environment (SE), physical environment (PE) and emulated environment (EE). Table 3 shows the classification of the environments, as described in each paper regarding the three categories . The network emulator Mininet 4 was the predominant system in all the works that use emulated environments. Table 4 shows details of the software configuration of the experimental environments discussed in each study. In a few cases, the authors clearly indicated the use of virtual machines (VMs) and we check marked those cases where the authors did not state the name or the version of the VMs. With regard to the operating systems (OS), we included the name and version described in the experiments. In the switch column, we showed the number of switches used in the experiments, followed by the model of the switch (virtual/software or physical). The last column shows the number of hosts used in the experiments. All the blank cells refer to studies in which the related item was not described.

Experimental Security Analysis of Controller Software in SDNs: A Review <s> Controller <s> This paper presents an independent comprehensive analysis of the efficiency indexes of popular open source SDN/OpenFlow controllers (NOX, POX, Beacon, Floodlight, MuL, Maestro, Ryu). The analysed indexes include performance, scalability, reliability, and security. For testing purposes we developed the new framework called hcprobe. The test bed and the methodology we used are discussed in detail so that everyone could reproduce our experiments. The result of the evaluation show that modern SDN/OpenFlow controllers are not ready to be used in production and have to be improved in order to increase all above mentioned characteristics. <s> BIB001 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Controller <s> Software Defined Networking (SDN) has been proposed as a drastic shift in the networking paradigm, by decoupling network control from the data plane and making the switching infrastructure truly programmable. The key enabler of SDN, OpenFlow, has seen widespread deployment on production networks and its adoption is constantly increasing. Although openness and programmability are primary features of OpenFlow, security is of core importance for real-world deployment. In this work, we perform a security analysis of OpenFlow using STRIDE and attack tree modeling methods, and we evaluate our approach on an emulated network testbed. The evaluation assumes an attacker model with access to the network data plane. Finally, we propose appropriate counter-measures that can potentially mitigate the security issues associated with OpenFlow networks. Our analysis and evaluation approach are not exhaustive, but are intended to be adaptable and extensible to new versions and deployment contexts of OpenFlow. <s> BIB002 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Controller <s> In this paper, for the first time we show a new attack to fin- gerprint SDN networks and further launch efficient resource consumption attacks. This attack demonstrates that SDN brings new security issues that may not be ignored. We provide the first feasibility study of such attack and hope to stimulate further studies in SDN security research. <s> BIB003 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Controller <s> Software-Defined Networking proposes a new paradigm by separating the control plane from the data plane. The control plane's main responsibility is deployed in a centralized component called SDN controller. Therefore, the SDN based network becomes dependent on a single functional entity (although it may have multiple instances) in order to accomplish its routing and policy rules. The SDN controller is vulnerable to degradation due to a well-known phenomenon called Software Aging. To the best of our knowledge, this work is the first that assesses the impacts of software aging on current implementation of SDN controllers. We show that such a phenomenon must not be neglected in future software architecture designs and implementations. <s> BIB004 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Controller <s> One of the core properties of Software Defined Networking (SDN) is the ability for third parties to develop network applications. This introduces increased potential for innovation in networking from performance-enhanced to energy-efficient designs. In SDN, the application connects with the network via the SDN controller. A specific concern relating to this communication channel is whether an application can be trusted or not. For example, what information about the network state is gathered by the application? Is this information necessary for the application to execute or is it gathered for malicious intent? In this paper we present an approach to secure the northbound interface by introducing a permissions system that ensures that controller operations are available to trusted applications only. Implementation of this permissions system with our Operation Checkpoint adds negligible overhead and illustrates successful defense against unauthorized control function access attempts. <s> BIB005 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Controller <s> Within the hierarchy of the Software Defined Network (SDN) network stack, the control layer operates as the critical middleware facilitator of interactions between the data plane and the network applications, which govern flow routing decisions. In the OpenFlow implementation of the SDN model, the control layer, commonly referred to as a network operating system (NOS), has been realized by a range of competing implementations that offer various performance and functionality advantages: Floodlight, POX, NOX, and ONIX. In this paper we focus on the question of control layer resilience, when rapidly developed prototype network applications go awry, or third-party network applications incorporate unexpected vulnerabilities, fatal instabilities, or even malicious logic. We demonstrate how simple and common failures in a network application may lead to loss of the control layer, and in effect, loss of network control. To address these concerns we present the ROSEMARY controller, which implements a network application containment and resilience strategy based around the notion of spawning applications independently within a micro-NOS. ROSEMARY distinguishes itself by its blend of process containment, resource utilization monitoring, and an application permission structure, all designed to prevent common failures of network applications from halting operation of the SDN Stack. We present our design and implementation of ROSEMARY, along with an extensive evaluation of its performance relative to several of the mostly well-known and widely used controllers. Rather than imposing significant performance costs, we find that with the integration of two optimization features, ROSEMARY offers a competitive performance advantage over the majority of other controllers. <s> BIB006 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Controller <s> Both academic researchers and network vendors have started to promote Software-Defined Networking (SDN) as a new network paradigm, in which controller systems play a major role. A modular and extensible design allows network operators to extend the controller's functionality by so called network services. Unfortunately, in current designs such network services have unlimited access to mandatory SDN resources which enables different kinds of attacks. To retain control over network services (especially third-party ones), we adapt approved security mechanisms and propose a containment mechanism as well as a framework to ease containment configuration. For both proposals, we provide proof-of-concept implementations for an open and industry-supported reference framework and hereby aim to improve security for a wide range of SDN controllers. Finally, our proposals achieve the ability to harden a mandatory SDN component (i. e., the SDN controller) and enable proactive security even against malicious network services. <s> BIB007 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Controller <s> A lot of new features and capabilities emerge in the network because of separation of data plane and control plane in Software Defined Network (SDN) terminology. One of them is the possibility to implement Northbound API to allow third-party application to access resources of the network. However, most of the current implementations of Northbound API do not consider the security aspect. Therefore, we design more secure scheme for it. The design consists of token authentication for application and user, who is responsible to control/use the application/network, using OAuth 2.0 protocol. <s> BIB008 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Controller <s> Software-defined networks (SDNs) allow greater control over network entities by centralizing the control plane, but place great burden on the administrator to manually ensure security and correct functioning of the entire network. We list several attacks on SDN controllers that violate network topology and data plane forwarding, and can be mounted by compromised network entities, such as end hosts and soft switches. We further demonstrate their feasibility on four popular SDN controllers. We propose SPHINX to detect both known and potentially unknown attacks on network topology and data plane forwarding originating within an SDN. SPHINX leverages the novel abstraction of flow graphs, which closely approximate the actual network operations, to enable incremental validation of all network updates and constraints. SPHINX dynamically learns new network behavior and raises alerts when it detects suspicious changes to existing network control plane behavior. Our evaluation shows that SPHINX is capable of detecting attacks in SDNs in realtime with low performance overheads, and requires no changes to the controllers for deployment. <s> BIB009 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Controller <s> The SDN paradigm allows networks to be dynamically reconfigurable by network applications. SDN is also of particular interest for NFV which deals with the virtualization of network functions. The network programmability offered by SDN presents then various advantages but it also induces various threats regarding potential attacks on the network. For instance, there is a critical risk that a hacker takes over the network control by exploiting this SDN network programmability (e.g., using the SDN API or tampering a network application running on the SDN controller). This paper proposes then an approach to deal with this possible lack of trust in the SDN controller or in their applications. This approach consists in not relying on a single controller but on several ‘redundant’ controllers that may also run in different execution environments. The network configuration requests coming from these controllers are then compared and, if deemed sufficiently consistent and then trustable, they are actually sent to the network. This approach has been implemented in an intermediary layer (based on a network hypervisor) inserted between the network equipments and the controllers. Experimentations have been performed showing the feasibility of the approach and providing some first evaluations of its impact on the network and the services. <s> BIB010 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Controller <s> Topology Discovery is an essential service in Software Defined Networks (SDN). Most SDN controllers use a de-facto standard topology discovery mechanism based on Open-Flow to identify active links in the network. This paper discusses the security, or rather lack thereof, of the current SDN topology discovery mechanism, and its vulnerability to link spoofing attacks. The feasibility and impact of the attacks are verified and demonstrated via experiments. The paper presents and evaluates a countermeasure based on HMAC authentication. <s> BIB011 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Controller <s> Abstract Logically centralized nature of the controller in Software-defined Networking (SDN) makes it vulnerable to various adversarial attacks. These attacks have capabilities to degrade the performance of the managed network or bring it down in the worst case. For this reason, a large-scale deployment of SDN needs evaluation of the impact of adversarial attacks on network services. In this work, we implement various attacks in SDN and analyze their impact on the performance of web-based services running over it. Prior to that, we briefly discuss different kinds of vulnerabilities and threats in SDN. We consider the connection set-up latency and loss, for web-client requests, as metrics for the evaluation. We observe a significant degradation in the performances of web services, in terms of response time and availability, in the presence of implemented attacks. <s> BIB012 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Controller <s> In a virtualized environment, different virtual networks can operate over the same physical infrastructure. Each virtual network has its own protocols and share the available resources, thus highlighting the need of resource isolation mechanisms. Investigating the isolation mechanisms provided by FlowVisor, we have discovered vulnerabilities previously unknown regarding addressing space isolation. We show that, in the presence of a malicious controller, FlowVisor’s isolation can be broken allowing different attacks. This paper addresses these vulnerabilities by proposing an Action Slicing mechanism, that allows FlowVisor to limit which actions can be used by each virtual network controller, thus extending the virtual network definition. Our experimental results show that using the proposed Action Slicing mechanism can effectively neutralize the discovered vulnerabilities. <s> BIB013 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Controller <s> In this paper, we investigated the vulnerabilities surrounding software-defined networking (SDN). Specifically, we examined the vulnerabilities of OpenDayLight SDN controller. Among all the possible attack vectors, a man-in-the-middle attack using ARP poisoning was successfully launched to intercept the traffic between a client and the OpenDayLight controller. Details of the experiment method, procedures and results were described in this manuscript. To the best of our knowledge, this is the first successful practical attempt to penetrate an SDN controller and be able to capture login credentials of the controller. The significance of this attack should not be taken lightly; once the SDN controller is under the control of the adversary, there will be no security at all for the entire network governed by this controller. <s> BIB014 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Controller <s> SDN is a concept of programmable networking. A network manager can process each network flow through software programs. There is a hypothesis that all switches are trusted and follow programmed commands to handle packets. That is, once a switch is compromised by an attacker and does not follow the order of the network manager, this will bring a huge network disaster. In this paper, we define some attack models through compromised switches and design a detection mechanism to find these compromised devices. We evaluate our mechanism and discuss some future works. <s> BIB015 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Controller <s> Software-defined networking (SDN) enables the exible and dynamic configuration of a network, and OpenFlow is one practical SDN implementation. Although it has been widely deployed in actual environments, it can cause fatal aws. In this paper, we consolidate the security threats to OpenFlow mentioned in previous work and introduce a new security checksheet that includes risk assessment methods. We compare the Kreutz et al. threat vectors with the SDNSecurity.org attack list to discover new threats. Our checksheet enables the security of a given OpenFlow network design to be comprehensively assessed. Furthermore, we evaluate the performance of an OpenFlow network with two attack scenarios using the checksheet and identify critical performance degradations. <s> BIB016 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Controller <s> Software-Defined Networking (SDN) allows scalable and flexible network management without requiring expensive hardware changes. However, this technology is relatively new and creates novel security risks. In this paper, we investigate the vulnerability of link discovery service in SDN controller, which is a critical service provided by the control layer for the proper functioning of applications and network services. We also discuss the potential attacks on link discovery service, which are verified via experiments. <s> BIB017 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Controller <s> Software Defined Networking (SDN) provides a simpler and more efficient approach to manage computer networks. This is achieved by decoupling the control functionality from the forwarding elements, and placing it into a logically centralised controller. This centralised architecture makes SDN susceptible to a range of Denial of Service (DoS) attacks. In this paper we investigate a range of DoS attacks against OpenFlow-based SDN, both against the control plane and data plane, and quantify the impact of the attacks via experiments. In particular, we compare the impact on three key SDN controller platforms, i.e. ONOS, Floodlight and Ryu. Our results show that with relatively limited resources, an attacker can cause a significant disruption against an SDN. <s> BIB018

Papers Amount FloodLight , BIB004 , BIB005 , BIB007 , BIB008 , BIB001 , BIB006 , , , BIB009 , , , BIB010 , BIB017 , BIB018 15 POX BIB002 , BIB008 , BIB011 , BIB001 , BIB006 , BIB012 , BIB013 , BIB009 , 9 OpenDayLight , BIB007 , BIB008 , BIB014 , BIB006 , , BIB009 , 8 Beacon BIB004 , BIB007 , BIB001 , BIB003 , BIB006 , 6 RYU BIB015 , BIB008 , BIB001 , BIB016 , , [ It should be noted that BIB017 did not specify any of the software used in their experiments. The overview, obtained from this data compilation, may support the choice for setups of SDN testbeds in future works, as well as providing the metrics and parameters required for qualitative evaluations between experimental platforms.

Experimental Security Analysis of Controller Software in SDNs: A Review <s> Security Threat Modeling and Classification <s> In today's hostile online environment, software must be designed to withstand malicious attacks of all kinds. Unfortunately, even security-conscious products can fall prey when designers fail to understand the threats their software faces or the ways in which adversaries might try to attack it. To better understand a product's threat environment and defend against potential attacks, Microsoft uses threat modeling, which should be treated like any other part of the design and specification process. In fact, singling it out as a special activity performed outside the normal design process actually detracts from its importance to the overall development life cycle. We must consider security needs throughout the design process, just as we do with performance, usability, localizability, serviceability, or any other facet. <s> BIB001 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> Security Threat Modeling and Classification <s> Software Defined Networking (SDN) has been proposed as a drastic shift in the networking paradigm, by decoupling network control from the data plane and making the switching infrastructure truly programmable. The key enabler of SDN, OpenFlow, has seen widespread deployment on production networks and its adoption is constantly increasing. Although openness and programmability are primary features of OpenFlow, security is of core importance for real-world deployment. In this work, we perform a security analysis of OpenFlow using STRIDE and attack tree modeling methods, and we evaluate our approach on an emulated network testbed. The evaluation assumes an attacker model with access to the network data plane. Finally, we propose appropriate counter-measures that can potentially mitigate the security issues associated with OpenFlow networks. Our analysis and evaluation approach are not exhaustive, but are intended to be adaptable and extensible to new versions and deployment contexts of OpenFlow. <s> BIB002

In the set of papers, we analyzed the Microsoft model for security threat classification (STRIDE) and Data Flow Diagrams (DFDs) appear as the most widely used threat modeling techniques. STRIDE, allows the security of a system to be classified into six categories: Spoofing, Tampering, Repudiation, Information disclosure, Denial of service, and Elevation of privileges. With the aid of DFDs, the security analyst is able to make a graphical representation of the software components, the inputs and outputs, as well as the internal logical processes BIB001 . Besides STRIDE and DFD, there are other alternative methods to assist the modeling of threats and security vulnerabilities. One of them is the Attack Tree, which is able to classify the ways needed to detect a security breach. The last method, the Petri Net, is a classic mathematical modeling language. ------ Figure 4 illustrates the trend in the use of threat modeling techniques, as revealed in the number of published papers, during the years. Figure 5 shows trends regarding the interest in each STRIDE category, reflected in the number of papers addressing them. The greatest interest was displayed in the Denial of Service attack types since these can either lead to the network malfunction or collapse. However, there is also an increasing concern about spoofing threats. With respect to security properties, Figure 6 illustrates the trend in the area of research, by showing the number of papers referring each security property. It should be noted that the terms "authentication" and "availability" are mentioned in most of the papers we analyzed. This is consistent with the occurrence of the related terms in the STRIDE model. However, it is also clear that authorization has also attracted greater attention in the last two years. According to the review carried out, the STRIDE model is widely adopted and is currently the standard method for categorizing security threats BIB002 . We set out guidelines in our literature review with regard to security threats, by classifying the papers described in Section 3 based on the STRIDE model, as shown in Table 5 .

Experimental Security Analysis of Controller Software in SDNs: A Review <s> ONF's Security Requirements for SDN Controllers <s> This paper presents an independent comprehensive analysis of the efficiency indexes of popular open source SDN/OpenFlow controllers (NOX, POX, Beacon, Floodlight, MuL, Maestro, Ryu). The analysed indexes include performance, scalability, reliability, and security. For testing purposes we developed the new framework called hcprobe. The test bed and the methodology we used are discussed in detail so that everyone could reproduce our experiments. The result of the evaluation show that modern SDN/OpenFlow controllers are not ready to be used in production and have to be improved in order to increase all above mentioned characteristics. <s> BIB001 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> ONF's Security Requirements for SDN Controllers <s> Software defined networking (SDN) decouples the network control and data planes. The network intelligence and state are logically centralized and the underlying network infrastructure is abstracted from applications. SDN enhances network security by means of global visibility of the network state where a conflict can be easily resolved from the logically centralized control plane. Hence, the SDN architecture empowers networks to actively monitor traffic and diagnose threats to facilitates network forensics, security policy alteration, and security service insertion. The separation of the control and data planes, however, opens security challenges, such as man-in-the middle attacks, denial of service (DoS) attacks, and saturation attacks. In this paper, we analyze security threats to application, control, and data planes of SDN. The security platforms that secure each of the planes are described followed by various security approaches for network-wide security in SDN. SDN security is analyzed according to security dimensions of the ITU-T recommendation, as well as, by the costs of security solutions. In a nutshell, this paper highlights the present and future security challenges in SDN and future directions for secure SDN. <s> BIB002 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> ONF's Security Requirements for SDN Controllers <s> SDN is a concept of programmable networking. A network manager can process each network flow through software programs. There is a hypothesis that all switches are trusted and follow programmed commands to handle packets. That is, once a switch is compromised by an attacker and does not follow the order of the network manager, this will bring a huge network disaster. In this paper, we define some attack models through compromised switches and design a detection mechanism to find these compromised devices. We evaluate our mechanism and discuss some future works. <s> BIB003 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> ONF's Security Requirements for SDN Controllers <s> Topology Discovery is an essential service in Software Defined Networks (SDN). Most SDN controllers use a de-facto standard topology discovery mechanism based on Open-Flow to identify active links in the network. This paper discusses the security, or rather lack thereof, of the current SDN topology discovery mechanism, and its vulnerability to link spoofing attacks. The feasibility and impact of the attacks are verified and demonstrated via experiments. The paper presents and evaluates a countermeasure based on HMAC authentication. <s> BIB004 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> ONF's Security Requirements for SDN Controllers <s> In recent years, Software-Defined Networking (SDN) has been a focus of research. As a promising network architecture, SDN will possibly replace traditional networking, as it brings promising opportunities for network management in terms of simplicity, programmability, and elasticity. While many efforts are currently being made to standardize this emerging paradigm, careful attention needs to be also paid to security at this early design stage. This paper focuses on the security aspects of SDN. We begin by discussing characteristics and standards of SDN. On the basis of these, we discuss the security features as a whole and then analyze the security threats and countermeasures in detail from three aspects, based on which part of the SDN paradigm they target, i.e., the data forwarding layer, the control layer and the application layer. Countermeasure techniques that could be used to prevent, mitigate, or recover from some of such attacks are also described, while the threats encountered when developing these defensive mechanisms are highlighted. <s> BIB005 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> ONF's Security Requirements for SDN Controllers <s> The revolutionary concept of Software Defined Networks (SDNs) potentially provides flexible and well-managed next-generation networks. All the hype surrounding the SDNs is predominantly because of its centralized management functionality, the separation of the control plane from the data forwarding plane, and enabling innovation through network programmability. Despite the promising architecture of SDNs, security was not considered as part of the initial design. Moreover, security concerns are potentially augmented considering the logical centralization of network intelligence. Furthermore, the security and dependability of the SDN has largely been a neglected topic and remains an open issue. The paper presents a broad overview of the security implications of each SDN layer/interface. This paper contributes further by devising a contemporary layered/interface taxonomy of the reported security vulnerabilities, attacks, and challenges of SDN. We also highlight and analyze the possible threats on each layer/interface of SDN to help design secure SDNs. Moreover, the ensuing paper contributes by presenting the state-of-the-art SDNs security solutions. The categorization of solutions is followed by a critical analysis and discussion to devise a comprehensive thematic taxonomy. We advocate the production of secure and dependable SDNs by presenting potential requirements and key enablers. Finally, in an effort to anticipate secure and dependable SDNs, we present the ongoing open security issues, challenges and future research directions. <s> BIB006 </s> Experimental Security Analysis of Controller Software in SDNs: A Review <s> ONF's Security Requirements for SDN Controllers <s> Due to the widespread research on Software Defined Networks (SDNs), its security has received much attention recently. But most of those attempts consider SDN security from the OpenFlow perspective. To the best of our knowledge, none so far has paid attention to the security analysis and modeling of Forwarding and Control planes Separation Network Structure (FCSNS) in SDN. Therefore, this paper provides a different approach to network security based on Petri net and Attack tree models. Our objective is to analyze the FCSNS security via the combination of model and state. This method represents the network structure and state transferring by way of Petri net. In addition, it introduces the security analysis method of STRIDE to build up the Attack tree model. Finally, we analyze FCSNS via the combination of Petri net and Attack tree model and present the results. Our results are very promising in using such models to achieve such security objectives. <s> BIB007

To the best of our knowledge, the recommendations of ONF are the first known security requirements defined for SDN controllers. There are several approaches in the literature that deal with network security in SDN, as discussed in the previous sections. However, they mostly focus on a single vulnerability or threat. The requirements defined in are important because they provide the basis of a security analysis by stressing the need to evaluate several points of vulnerability. Several surveys have been carried out with focus on the analysis of security in SDNs BIB005 , BIB002 , BIB006 , BIB001 , . However, the ONF report is a technical document containing a comprehensive practical coverage of security requirements for SDN controllers. Although a security evaluation of SND controllers is presented in , it is restricted to the analysis of five controllers for a subset of the ONF requirements. In light of this, there are open research issues for further investigation on the security of SDN controllers considering two scopes: in breadth, given that there is a wide diversity of controllers; and in depth, by taking into account all the ONF requirements. Table 6 lists the analyzed studies which address both the general and specific ONF security requirements . On the one hand, it was noted that some requirements have attracted more interest among researchers, (in Table 6 items Hardware consistency -0 5 Hypervisor security -0 6 Software package integrity -0 7 Protecting the integrity of data in transit BIB003 , BIB007 , BIB004 , 7, 9 of General Sec. Req. and items 6.a and 6.b of Specific Sec. Req.). On the other hand, several requirements were not addressed by any study (in Table 6 items 4, 5, 6, 8, 10, 12, 13, 14 and 18 of General Sec. Req. and item 4 of Specific Sec. Req.). Among the papers we examined, only does not fit into any ONF requirement. Since none of the requirements directly addresses the security in SDN switches, the study was not included in Table 6 .

Deep Reinforcement Learning for Autonomous Driving: A Survey <s> I. INTRODUCTION <s> Reinforcement learning, one of the most active research areas in artificial intelligence, is a computational approach to learning whereby an agent tries to maximize the total amount of reward it receives when interacting with a complex, uncertain environment. In Reinforcement Learning, Richard Sutton and Andrew Barto provide a clear and simple account of the key ideas and algorithms of reinforcement learning. Their discussion ranges from the history of the field's intellectual foundations to the most recent developments and applications. The only necessary mathematical background is familiarity with elementary concepts of probability. The book is divided into three parts. Part I defines the reinforcement learning problem in terms of Markov decision processes. Part II provides basic solution methods: dynamic programming, Monte Carlo methods, and temporal-difference learning. Part III presents a unified view of the solution methods and incorporates artificial neural networks, eligibility traces, and planning; the two final chapters present case studies and consider the future of reinforcement learning. <s> BIB001 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> I. INTRODUCTION <s> Deep Reinforcement Learning (DRL) has become increasingly powerful in recent years, with notable achievements such as Deepmind's AlphaGo. It has been successfully deployed in commercial vehicles like Mobileye's path planning system. However, a vast majority of work on DRL is focused on toy examples in controlled synthetic car simulator environments such as TORCS and CARLA. In general, DRL is still at its infancy in terms of usability in real-world applications. Our goal in this paper is to encourage real-world deployment of DRL in various autonomous driving (AD) applications. We first provide an overview of the tasks in autonomous driving systems, reinforcement learning algorithms and applications of DRL to AD systems. We then discuss the challenges which must be addressed to enable further progress towards real-world deployment. <s> BIB002

Autonomous driving (AD) 1 systems constitute of multiple perception level tasks that have now achieved high precision on account of deep learning architectures. Besides the perception, autonomous driving systems constitute of multiple tasks where classical supervised learning methods are no more applicable. First, when the prediction of the agent's action changes future sensor observations received from the environment under which the autonomous driving agent operates, for example the task of optimal driving speed in an urban area. Second, supervisory signals such as time to collision (TTC), lateral error w.r.t to optimal trajectory of the agent, represent the dynamics of the agent, as well uncertainty in the environment. Such problems would require defining the stochastic cost function to be maximized. Third, the agent is required to learn new configurations of the environment, as well as to predict an optimal decision at each instant while driving in its environment. This represents a high dimensional space given the number of unique configurations under which the agent & environment are observed, this is combinatorially large. In all such scenarios we are aiming to solve a sequential decision process, which is formalized under the classical settings of Reinforcement Learning (RL), where the agent is required to learn and represent its environment as well as BIB001 (Tables III and IV). act optimally given at each instant BIB001 . The optimal action is referred to as the policy. In this review we shall cover the notions of reinforcement learning, the taxonomy of tasks where RL is a promising solution especially in the domains of driving policy, predictive perception, path and motion planning, and low level controller design. We also focus our review on the different real world deployments of RL in the domain of autonomous driving expanding our conference paper BIB002 since their deployment has not been reviewed in an academic setting. Finally, we motivate users by demonstrating the key computational challenges and risks when applying current day RL algorithms such imitation learning, deep Q learning, among others. The rest of the paper is organized as follows. Section II provides an overview of components of a typical autonomous driving system. Section III provides an introduction to reinforcement learning and briefly discusses key concepts. Section IV discusses more sophisticated extensions on top of the basic RL framework. Section V provides an overview of RL applications for autonomous driving problems. Section VI discusses challenges in deploying RL for real-world autonomous driving systems. Section VII concludes this paper with some final remarks. Figure 1 comprises of the standard blocks of an AD system demonstrating the pipeline from sensor stream to control actuation. The sensor architecture in a modern autonomous driving system notably includes multiple sets of cameras, radars and LIDARs as well as a GPS-GNSS system for absolute localisation and inertial measurement Units (IMUs) that provide 3D pose of the vehicle in space.

Deep Reinforcement Learning for Autonomous Driving: A Survey <s> II. COMPONENTS OF AD SYSTEM <s> Semantic segmentation was seen as a challenging computer vision problem few years ago. Due to recent advancements in deep learning, relatively accurate solutions are now possible for its use in automated driving. In this paper, the semantic segmentation problem is explored from the perspective of automated driving. Most of the current semantic segmentation algorithms are designed for generic images and do not incorporate prior structure and end goal for automated driving. First, the paper begins with a generic taxonomic survey of semantic segmentation algorithms and then discusses how it fits in the context of automated driving. Second, the particular challenges of deploying it into a safety system which needs high level of accuracy and robustness are listed. Third, different alternatives instead of using an independent semantic segmentation module are explored. Finally, an empirical evaluation of various semantic segmentation architectures was performed on CamVid dataset in terms of accuracy and speed. This paper is a preliminary shorter version of a more detailed survey which is work in progress. <s> BIB001 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> II. COMPONENTS OF AD SYSTEM <s> For autonomous driving, moving objects like vehicles and pedestrians are of critical importance as they primarily influence the maneuvering and braking of the car. Typically, they are detected by motion segmentation of dense optical flow augmented by a CNN based object detector for capturing semantics. In this paper, our aim is to jointly model motion and appearance cues in a single convolutional network. We propose a novel two-stream architecture for joint learning of object detection and motion segmentation. We designed three different flavors of our network to establish systematic comparison. It is shown that the joint training of tasks significantly improves accuracy compared to training them independently. Although motion segmentation has relatively fewer data than vehicle detection. The shared fusion encoder benefits from the joint training to learn a generalized representation. We created our own publicly available dataset (KITTI MOD) by extending KITTI object detection to obtain static/moving annotations on the vehicles. We compared against MPNet as a baseline, which is the current state of the art for CNN-based motion detection. It is shown that the proposed two-stream architecture improves the mAP score by 21.5% in KITTI MOD. We also evaluated our algorithm on the non-automotive DAVIS dataset and obtained accuracy close to the state-of-the-art performance. The proposed network runs at 8 fps on a Titan X GPU using a basic VGG16 encoder. <s> BIB002 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> II. COMPONENTS OF AD SYSTEM <s> Near-field depth estimation around a self-driving car is an important function that can be achieved by four wide-angle fisheye cameras having a field of view of over 180°. Depth estimation based on convolutional neural networks (CNNs) produce state of the art results, but progress is hindered because depth annotation cannot be obtained manually. Synthetic datasets are commonly used but they have limitations. For instance, they do not capture the extensive variability in the appearance of objects like vehicles present in real datasets. There is also a domain shift while performing inference on natural images illustrated by many attempts to handle the domain adaptation explicitly. In this work, we explore an alternate approach of training using sparse LiDAR data as ground truth for depth estimation for fisheye camera. We built our own dataset using our self-driving car setup which has a 64-beam Velodyne LiDAR and four wide angle fisheye cameras. To handle the difference in view-points of LiDAR and fisheye camera, an occlusion resolution mechanism was implemented. We started with Eigen's multiscale convolutional network architecture [1] and improved by modifying activation function and optimizer. We obtained promising results on our dataset with RMSE errors comparable to the state-of-the-art results obtained on KITTI. <s> BIB003 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> II. COMPONENTS OF AD SYSTEM <s> LiDAR has become a standard sensor for autonomous driving applications as they provide highly precise 3D point clouds. LiDAR is also robust for low-light scenarios at night-time or due to shadows where the performance of cameras is degraded. LiDAR perception is gradually becoming mature for algorithms including object detection and SLAM. However, semantic segmentation algorithm remains to be relatively less explored. Motivated by the fact that semantic segmentation is a mature algorithm on image data, we explore sensor fusion based 3D segmentation. Our main contribution is to convert the RGB image to a polar-grid mapping representation used for LiDAR and design early and mid-level fusion architectures. Additionally, we design a hybrid fusion architecture that combines both fusion algorithms. We evaluate our algorithm on KITTI dataset which provides segmentation annotation for cars, pedestrians and cyclists. We evaluate two state-of-the-art architectures namely SqueezeSeg and PointSeg and improve the mIoU score by 10% in both cases relative to the LiDAR only baseline. <s> BIB004 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> II. COMPONENTS OF AD SYSTEM <s> Cameras are an essential part of sensor suite in autonomous driving. Surround-view cameras are directly exposed to external environment and are vulnerable to get soiled. Cameras have a much higher degradation in performance due to soiling compared to other sensors. Thus it is critical to accurately detect soiling on the cameras, particularly for higher levels of autonomous driving. We created a new dataset having multiple types of soiling namely opaque and transparent. It will be released publicly as part of our WoodScape dataset [15] to encourage further research. We demonstrate high accuracy using a Convolutional Neural Network (CNN) based architecture. We also show that it can be combined with the existing object detection task in a multi-task learning framework. Finally, we make use of Generative Adversarial Networks (GANs) to generate more images for data augmentation and show that it works successfully similar to the style transfer. <s> BIB005 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> II. COMPONENTS OF AD SYSTEM <s> Convolutional Neural Networks (CNNs) are successfully used for the important automotive visual perception tasks including object recognition, motion and depth estimation, visual SLAM, etc. However, these tasks are typically independently explored and modeled. In this paper, we propose a joint multi-task network design for learning several tasks simultaneously. Our main motivation is the computational efficiency achieved by sharing the expensive initial convolutional layers between all tasks. Indeed, the main bottleneck in automated driving systems is the limited processing power available on deployment hardware. There is also some evidence for other benefits in improving accuracy for some tasks and easing development effort. It also offers scalability to add more tasks leveraging existing features and achieving better generalization. We survey various CNN based solutions for visual perception tasks in automated driving. Then we propose a unified CNN model for the important tasks and discuss several advanced optimization and architecture design techniques to improve the baseline model. The paper is partly review and partly positional with demonstration of several preliminary results promising for future research. We first demonstrate results of multi-stream learning and auxiliary learning which are important ingredients to scale to a large multi-task model. Finally, we implement a two-stream three-task network which performs better in many cases compared to their corresponding single-task models, while maintaining network size. <s> BIB006 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> II. COMPONENTS OF AD SYSTEM <s> Fisheye cameras are commonly employed for obtaining a large field of view in surveillance, augmented reality and in particular automotive applications. In spite of their prevalence, there are few public datasets for detailed evaluation of computer vision algorithms on fisheye images. We release the first extensive fisheye automotive dataset, WoodScape, named after Robert Wood who invented the fisheye camera in 1906. WoodScape comprises of four surround view cameras and nine tasks including segmentation, depth estimation, 3D bounding box detection and soiling detection. Semantic annotation of 40 classes at the instance level is provided for over 10,000 images and annotation for other tasks are provided for over 100,000 images. With WoodScape, we would like to encourage the community to adapt computer vision models for fisheye camera instead of using naive rectification. <s> BIB007

The goal of the perception module is the creation of an intermediate level representation of the environment state (for example bird-eye view map of all obstacles and agents) that is to be later utilised by a decision making system that ultimately produces the driving policy. This state would include lane position, drivable zone, location of agents such as cars & pedestrians, state of traffic lights and others. Uncertainties in the perception propagate to the rest of the information chain. Robust sensing is critical for safety thus using redundant sources increases confidence in detection. This is achieved by a combination of several perception tasks like semantic segmentation BIB001 , BIB004 , motion estimation BIB002 , depth estimation BIB003 , soiling detection BIB005 , etc which can be efficiently unified into a multi-task model BIB006 , BIB007 .

Deep Reinforcement Learning for Autonomous Driving: A Survey <s> D. Control <s> This paper describes a real-time motion planning algorithm, based on the rapidly-exploring random tree (RRT) approach, applicable to autonomous vehicles operating in an urban environment. Extensions to the standard RRT are predominantly motivated by: 1) the need to generate dynamically feasible plans in real-time; 2) safety requirements; 3) the constraints dictated by the uncertain operating (urban) environment. The primary novelty is in the use of closed-loop prediction in the framework of RRT. The proposed algorithm was at the core of the planning and control software for Team MIT's entry for the 2007 DARPA Urban Challenge, where the vehicle demonstrated the ability to complete a 60 mile simulated military supply mission, while safely interacting with other autonomous and human driven vehicles. <s> BIB001 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> D. Control <s> Self-driving vehicles are a maturing technology with the potential to reshape mobility by enhancing the safety, accessibility, efficiency, and convenience of automotive transportation. Safety-critical tasks that must be executed by a self-driving vehicle include planning of motions through a dynamic environment shared with other vehicles and pedestrians, and their robust executions via feedback control. The objective of this paper is to survey the current state of the art on planning and control algorithms with particular regard to the urban setting. A selection of proposed techniques is reviewed along with a discussion of their effectiveness. The surveyed approaches differ in the vehicle mobility model used, in assumptions on the structure of the environment, and in computational requirements. The side-by-side comparison presented in this survey helps to gain insight into the strengths and limitations of the reviewed approaches and assists with system level design choices. <s> BIB002

A controller defines the speed, steering angle and braking actions necessary over every point in the path obtained from a pre-determined map such as Google maps, or expert driving recording of the same values at every waypoint. Trajectory tracking in contrast involves a temporal model of the dynamics of the vehicle viewing the waypoints sequentially over time. Current vehicle control methods are founded in classical optimal control theory which can be stated as a minimisation of a cost functionẋ = f (x(t), u(t)) defined over a set of states x(t) and control actions u(t). The control input is usually defined over a finite time horizon and restricted on a feasible state space x ∈ X free BIB001 . The velocity control are based on classical methods of closed loop control such as PID (proportionalintegral-derivative) controllers, MPC (Model predictive control). PIDs aim to minimise a cost function constituting of three terms current error with proportional term, effect of past errors with integral term, and effect of future errors with the derivative term. While the family of MPC methods aim to stabilize the behavior of the vehicle while tracking the specified path BIB002 . A review on controllers, motion planning and learning based approaches for the same are provided in this review for interested readers. Optimal control and reinforcement learning are intimately related, where optimal control can be viewed as a model based reinforcement learning problem where the dynamics of the vehicle/environment are modeled by well defined differential equations. Reinforcement learning methods were developed to handle stochastic control problems as well ill-posed problems with unknown rewards and state transition probabilities.

Deep Reinforcement Learning for Autonomous Driving: A Survey <s> III. REINFORCEMENT LEARNING <s> Reinforcement learning, one of the most active research areas in artificial intelligence, is a computational approach to learning whereby an agent tries to maximize the total amount of reward it receives when interacting with a complex, uncertain environment. In Reinforcement Learning, Richard Sutton and Andrew Barto provide a clear and simple account of the key ideas and algorithms of reinforcement learning. Their discussion ranges from the history of the field's intellectual foundations to the most recent developments and applications. The only necessary mathematical background is familiarity with elementary concepts of probability. The book is divided into three parts. Part I defines the reinforcement learning problem in terms of Markov decision processes. Part II provides basic solution methods: dynamic programming, Monte Carlo methods, and temporal-difference learning. Part III presents a unified view of the solution methods and incorporates artificial neural networks, eligibility traces, and planning; the two final chapters present case studies and consider the future of reinforcement learning. <s> BIB001 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> III. REINFORCEMENT LEARNING <s> Efficient exploration remains a challenging research problem in reinforcement learning, especially when an environment contains large state spaces, deceptive local optima, or sparse rewards. To tackle this problem, we present a diversity-driven approach for exploration, which can be easily combined with both off- and on-policy reinforcement learning algorithms. We show that by simply adding a distance measure to the loss function, the proposed methodology significantly enhances an agent's exploratory behaviors, and thus preventing the policy from being trapped in local optima. We further propose an adaptive scaling method for stabilizing the learning process. Our experimental results in Atari 2600 show that our method outperforms baseline approaches in several tasks in terms of mean scores and exploration efficiency. <s> BIB002

Machine learning (ML) is a process whereby a computer program learns from experience to improve its performance at a specified task . ML algorithms are often classified under one of three broad categories: supervised learning, unsupervised learning and reinforcement learning (RL). Supervised learning algorithms are based on inductive inference where the model is typically trained using labelled data to perform classification or regression, whereas unsupervised learning encompasses techniques such as density estimation or clustering applied to unlabelled data. By contrast, in the RL paradigm an autonomous agent learns to improve its performance at an assigned task by interacting with its environment. Russel and Norvig define an agent as "anything that can be viewed as perceiving its environment through sensors and acting upon that environment through actuators" . RL agents are not told explicitly how to act by an expert; rather an agent's performance is evaluated by a reward function R. For each state experienced, the agent chooses an action and receives an occasional reward from its environment based on the usefulness of its decision. The goal for the agent is to maximize the cumulative rewards received over its lifetime. Gradually, the agent can increase its long-term reward by exploiting knowledge learned about the expected utility (i.e. discounted sum of expected future rewards) of different stateaction pairs. One of the main challenges in reinforcement learning is managing the trade-off between exploration and exploitation. To maximize the rewards it receives, an agent must exploit its knowledge by selecting actions which are known to result in high rewards. On the other hand, to discover such beneficial actions, it has to take the risk of trying new actions which may lead to higher rewards than the current best-valued actions for each system state. In other words, the learning agent has to exploit what it already knows in order to obtain rewards, but it also has to explore the unknown in order to make better action selections in the future. Examples of strategies which have been proposed to manage this trade-off include -greedy and softmax. When adopting the ubiquitous -greedy strategy, an agent either selects an action at random with probability 0 < < 1, or greedily selects the highest valued action for the current state with the remaining probability 1 − . Intuitively, the agent should explore more at the beginning of the training process when little is known about the problem environment. As training progresses, the agent may gradually conduct more exploitation than exploration. The design of exploration strategies for RL agents is an area of active research (see e.g. BIB002 ). Markov decision processes (MDPs) are considered the de facto standard when formalising sequential decision making problems involving a single RL agent . An MDP consists of a set S of states, a set A of actions, a transition function T and a reward function R probability T(s, a, s ) ∈ (0, 1), and give a reward R(s, a). This process is illustrated in Fig. 2 . The stochastic policy π : S → D is a mapping from the state space to a probability over the set of actions, and π(a|s) represents the probability of choosing action a at state s. The goal is to find the optimal policy π * , which results in the highest expected sum of discounted rewards : for all states s ∈ S, where r k = R(s k , a k ) is the reward at time k and V π (s), the 'value function' at state s following a policy π, is the expected 'return' (or 'utility') when starting at s and following the policy π thereafter BIB001 . An important, related concept is the action-value function, a.k.a.'Q-function' defined as: The discount factor γ ∈ [0, 1] controls how an agent regards future rewards. Low values of γ encourage myopic behaviour where an agent will aim to maximise short term rewards, whereas high values of γ cause agents to be more forwardlooking and to maximise rewards over a longer time frame. The horizon H refers to the number of time steps in the MDP. In infinite-horizon problems H = ∞, whereas in episodic domains H has a finite value. Episodic domains may terminate after a fixed number of time steps, or when an agent reaches a specified goal state. The last state reached in an episodic domain is referred to as the terminal state. In finite-horizon or goal-oriented domains discount factors of (close to) 1 may be used to encourage agents to focus on achieving the goal, whereas in infinite-horizon domains lower discount factors may be used to strike a balance between short-and longterm rewards. If the optimal policy for a MDP is known, then V π * may be used to determine the maximum expected discounted sum of rewards available from any arbitrary initial state. A rollout is a trajectory produced in the state space by sequentially applying a policy to an initial state. A MDP satisfies the Markov property, i.e. system state transitions are dependent only on the most recent state and action, not on the full history of states and actions in the decision process. Moreover, in many real-world application domains, it is not possible for an agent to observe all features of the environment state; in such cases the decision-making problem is formulated as a partially-observable Markov decision process (POMDP). Solving a reinforcement learning task means finding a policy π that maximises the expected discounted sum of rewards over trajectories in the state space. RL agents may learn value function estimates, policies and/or environment models directly. Dynamic programming (DP) refers to a collection of algorithms that can be used to compute optimal policies given a perfect model of the environment in terms of reward and transition functions. Unlike DP, in Monte Carlo methods there is no assumption of complete environment knowledge. Monte Carlo methods are incremental in an episode-by-episode sense. Upon the completion of an episode, the value estimates and policies are updated. Temporal Difference (TD) methods, on the other hand, are incremental in a step-by-step sense, making them applicable to non-episodic scenarios. Like Monte Carlo methods, TD methods can learn directly from raw experience without a model of the environment's dynamics. Like DP, TD methods learn their estimates based on other estimates.

Deep Reinforcement Learning for Autonomous Driving: A Survey <s> A. Value-based methods <s> Q-learning (Watkins, 1989) is a simple way for agents to learn how to act optimally in controlled Markovian domains. It amounts to an incremental method for dynamic programming which imposes limited computational demands. It works by successively improving its evaluations of the quality of particular actions at particular states. This paper presents and proves in detail a convergence theorem for Q,-learning based on that outlined in Watkins (1989). We show that Q-learning converges to the optimum action-values with probability 1 so long as all actions are repeatedly sampled in all states and the action-values are represented discretely. We also sketch extensions to the cases of non-discounted, but absorbing, Markov environments, and where many Q values can be changed each iteration, rather than just one. <s> BIB001 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> A. Value-based methods <s> The theory of reinforcement learning provides a normative account, deeply rooted in psychological and neuroscientific perspectives on animal behaviour, of how agents may optimize their control of an environment. To use reinforcement learning successfully in situations approaching real-world complexity, however, agents are confronted with a difficult task: they must derive efficient representations of the environment from high-dimensional sensory inputs, and use these to generalize past experience to new situations. Remarkably, humans and other animals seem to solve this problem through a harmonious combination of reinforcement learning and hierarchical sensory processing systems, the former evidenced by a wealth of neural data revealing notable parallels between the phasic signals emitted by dopaminergic neurons and temporal difference reinforcement learning algorithms. While reinforcement learning agents have achieved some successes in a variety of domains, their applicability has previously been limited to domains in which useful features can be handcrafted, or to domains with fully observed, low-dimensional state spaces. Here we use recent advances in training deep neural networks to develop a novel artificial agent, termed a deep Q-network, that can learn successful policies directly from high-dimensional sensory inputs using end-to-end reinforcement learning. We tested this agent on the challenging domain of classic Atari 2600 games. We demonstrate that the deep Q-network agent, receiving only the pixels and the game score as inputs, was able to surpass the performance of all previous algorithms and achieve a level comparable to that of a professional human games tester across a set of 49 games, using the same algorithm, network architecture and hyperparameters. This work bridges the divide between high-dimensional sensory inputs and actions, resulting in the first artificial agent that is capable of learning to excel at a diverse array of challenging tasks. <s> BIB002

Q-learning is one of the most commonly used RL algorithms. It is a model-free TD algorithm that learns estimates of the utility of individual state-action pairs (Q-functions defined in Eqn. 2). Q-learning has been shown to converge to the optimum state-action values for a MDP with probability 1, so long as all actions in all states are sampled infinitely often and the state-action values are represented discretely BIB001 . In practice, Q-learning will learn (near) optimal stateaction values provided a sufficient number of samples are obtained for each state-action pair. If a Q-learning agent has converged to the optimal Q values for a MDP and selects actions greedily thereafter, it will receive the same expected sum of discounted rewards as calculated by the value function with π * (assuming that the same arbitrary initial starting state is used for both). Agents implementing Q-learning update their Q values according to the following update rule: where Q(s, a) is an estimate of the utility of selecting action a in state s, α ∈ [0, 1] is the learning rate which controls the degree to which Q values are updated at each time step, and γ ∈ [0, 1] is the same discount factor used in Eqn. 1. The theoretical guarantees of Q-learning hold with any arbitrary initial Q values BIB001 ; therefore the optimal Q values for a MDP can be learned by starting with any initial action value function estimate. The initialisation can be optimistic (each Q(s, a) returns the maximum possible reward), pessimistic (minimum) or even using knowledge of the problem to ensure faster convergence. Deep Q-Networks (DQN) BIB002 incorporates a variant of the Q-learning algorithm , by using deep neural networks (DNNs) as a non-linear Q function approximator over high-dimensional state spaces (e.g. the pixels in a frame of an Atari game). Practically, the neural network predicts the value of all actions without the use of any explicit domainspecific information or hand-designed features. DQN applies experience replay technique to break the correlation between successive experience samples and also for better sample efficiency. For increased stability, two networks are used where the parameters of the target network for DQN are fixed for a number of iterations while updating the parameters of the online network. Readers are directed to sub-section III-E for a more detailed introduction to the use of DNNs in Deep RL.

Deep Reinforcement Learning for Autonomous Driving: A Survey <s> B. Policy-based methods <s> This article presents a general class of associative reinforcement learning algorithms for connectionist networks containing stochastic units. These algorithms, called REINFORCE algorithms, are shown to make weight adjustments in a direction that lies along the gradient of expected reinforcement in both immediate-reinforcement tasks and certain limited forms of delayed-reinforcement tasks, and they do this without explicitly computing gradient estimates or even storing information from which such estimates could be computed. Specific examples of such algorithms are presented, some of which bear a close relationship to certain existing algorithms while others are novel but potentially interesting in their own right. Also given are results that show how such algorithms can be naturally integrated with backpropagation. We close with a brief discussion of a number of additional issues surrounding the use of such algorithms, including what is known about their limiting behaviors as well as further considerations that might be used to help develop similar but potentially more powerful reinforcement learning algorithms. <s> BIB001 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> B. Policy-based methods <s> In this paper we consider deterministic policy gradient algorithms for reinforcement learning with continuous actions. The deterministic policy gradient has a particularly appealing form: it is the expected gradient of the action-value function. This simple form means that the deterministic policy gradient can be estimated much more efficiently than the usual stochastic policy gradient. To ensure adequate exploration, we introduce an off-policy actor-critic algorithm that learns a deterministic target policy from an exploratory behaviour policy. We demonstrate that deterministic policy gradient algorithms can significantly outperform their stochastic counterparts in high-dimensional action spaces. <s> BIB002 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> B. Policy-based methods <s> In this article, we describe a method for optimizing control policies, with guaranteed monotonic improvement. By making several approximations to the theoretically-justified scheme, we develop a practical algorithm, called Trust Region Policy Optimization (TRPO). This algorithm is effective for optimizing large nonlinear policies such as neural networks. Our experiments demonstrate its robust performance on a wide variety of tasks: learning simulated robotic swimming, hopping, and walking gaits; and playing Atari games using images of the screen as input. Despite its approximations that deviate from the theory, TRPO tends to give monotonic improvement, with little tuning of hyperparameters. <s> BIB003 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> B. Policy-based methods <s> We propose a new family of policy gradient methods for reinforcement learning, which alternate between sampling data through interaction with the environment, and optimizing a "surrogate" objective function using stochastic gradient ascent. Whereas standard policy gradient methods perform one gradient update per data sample, we propose a novel objective function that enables multiple epochs of minibatch updates. The new methods, which we call proximal policy optimization (PPO), have some of the benefits of trust region policy optimization (TRPO), but they are much simpler to implement, more general, and have better sample complexity (empirically). Our experiments test PPO on a collection of benchmark tasks, including simulated robotic locomotion and Atari game playing, and we show that PPO outperforms other online policy gradient methods, and overall strikes a favorable balance between sample complexity, simplicity, and wall-time. <s> BIB004

The difference between value-based and policy-based methods is essentially a matter of where the burden of optimality resides. Both method types must propose actions and evaluate the resulting behaviour, but while value-based methods focus on evaluating the optimal cumulative reward and have a policy follows the recommendations, policy-based methods aim to estimate the optimal policy directly, and the value is a secondary if calculated at all. Typically, a policy is parameterised as a neural network π θ . Policy gradient methods use gradient descent to estimate the parameters of the policy that maximise the expected reward. The result can be a stochastic policy where actions are selected by sampling, or a deterministic policy. Many real-world applications have continuous action spaces. Deterministic policy gradient (DPG) algorithms BIB002 [1] allow reinforcement learning in domains with continuous actions. Silver et al. BIB002 proved that a deterministic policy gradient exists for MDPs satisfying certain conditions, and that deterministic policy gradients have a simple model-free form that follows the gradient of the action-value function. As a result, instead of integrating over both state and action spaces in stochastic policy gradients, DPG integrates over the state space only leading to fewer samples in problems with large action spaces. To ensure sufficient exploration, actions are chosen using a stochastic policy, while learning a deterministic target policy. The REINFORCE BIB001 algorithm is a straight forward policy-based method. The discounted cumulative reward g t = H−1 k=0 γ k r k+t+1 at one time step is calculated by playing the entire episode, so no estimator is required for policy evaluation. The parameters are updated into the direction of the performance gradient: where α is the learning rate for a stable incremental update. Intuitively, we want to encourage state-action pairs that result in the best possible returns. Trust Region Policy Optimization (TRPO) BIB003 , works by preventing the updated policies from deviating too much from previous policies, thus reducing the chance of a bad update. TRPO optimises a surrogate objective function where the basic idea is to limit each policy gradient update as measured by the Kullback-Leibler (KL) divergence between the current and the new proposed policy. This method results in monotonic improvements in policy performance. While Proximal Policy Optimization (PPO) BIB004 proposed a clipped surrogate objective function by adding a penalty for having a too large policy change. Accordingly, PPO policy optimisation is simpler to implement, and has better sample complexity while ensuring the deviation from the previous policy is relatively small.

Deep Reinforcement Learning for Autonomous Driving: A Survey <s> C. Actor-critic methods <s> This article presents a general class of associative reinforcement learning algorithms for connectionist networks containing stochastic units. These algorithms, called REINFORCE algorithms, are shown to make weight adjustments in a direction that lies along the gradient of expected reinforcement in both immediate-reinforcement tasks and certain limited forms of delayed-reinforcement tasks, and they do this without explicitly computing gradient estimates or even storing information from which such estimates could be computed. Specific examples of such algorithms are presented, some of which bear a close relationship to certain existing algorithms while others are novel but potentially interesting in their own right. Also given are results that show how such algorithms can be naturally integrated with backpropagation. We close with a brief discussion of a number of additional issues surrounding the use of such algorithms, including what is known about their limiting behaviors as well as further considerations that might be used to help develop similar but potentially more powerful reinforcement learning algorithms. <s> BIB001 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> C. Actor-critic methods <s> Disclosed are methods, systems, and apparatus, including computer programs encoded on computer storage media, for training an actor neural network used to select actions to be performed by an agent interacting with an environment. One of the methods includes obtaining a minibatch of experience tuples; and updating current values of the parameters of the actor neural network, comprising: for each experience tuple in the minibatch: processing the training observation and the training action in the experience tuple using a critic neural network to determine a neural network output for the experience tuple, and determining a target neural network output for the experience tuple; updating current values of the parameters of the critic neural network using errors between the target neural network outputs and the neural network outputs; and updating the current values of the parameters of the actor neural network using the critic neural network. <s> BIB002

Actor-critic methods are hybrid methods that combine the benefits of policy-based and value-based algorithms. The policy structure that is responsible for selecting actions is known as the 'actor'. The estimated value function criticises the actions made by the actor and is known as the 'critic'. After each action selection, the critic evaluates the new state to determine whether the result of the selected action was better or worse than expected. Both networks need their gradient to learn. Let J(θ) := E π θ [r] represent a policy objective function, where θ designates the parameters of a DNN. Policy gradient methods search for local maximum of J(θ). Since optimization in continuous action spaces could be costly and slow, the DPG (Direct Policy Gradient) algorithm represents actions as parameterised function µ(s|θ µ ), where θ µ refers to the parameters of the actor network. Then the unbiased estimate of the policy gradient gradient step is given as: where b is the baseline. While using b ≡ 0 is the simplification that leads to the REINFORCE formulation. Williams BIB001 explains a well chosen baseline can reduce variance leading to a more stable learning. The baseline, b can be chosen as V π (s), Q π (s, a) or 'Advantage' A π (s, a) based methods. Deep Deterministic Policy Gradient (DDPG) BIB002 is a modelfree, off-policy (please refer to subsection III-D for a detailed distinction), actor-critic algorithm that can learn policies for continuous action spaces using deep neural net based function approximation, extending prior work on DPG to large and high-dimensional state-action spaces. When selecting actions, exploration is performed by adding noise to the actor policy. Like DQN, to stabilise learning a replay buffer is used to minimize data correlation. A separate actor-critic specific target network is also used. Normal Q-learning is adapted with a restricted number of discrete actions, and DDPG also needs a straightforward way to choose an action. Starting from Qlearning, we extend Eqn. 2 to define the optimal Q-value and optimal action as Q * and a * . a * = argmax a Q * (s, a).

Deep Reinforcement Learning for Autonomous Driving: A Survey <s> (7) <s> We propose a conceptually simple and lightweight framework for deep reinforcement learning that uses asynchronous gradient descent for optimization of deep neural network controllers. We present asynchronous variants of four standard reinforcement learning algorithms and show that parallel actor-learners have a stabilizing effect on training allowing all four methods to successfully train neural network controllers. The best performing method, an asynchronous variant of actor-critic, surpasses the current state-of-the-art on the Atari domain while training for half the time on a single multi-core CPU instead of a GPU. Furthermore, we show that asynchronous actor-critic succeeds on a wide variety of continuous motor control problems as well as on a new task of navigating random 3D mazes using a visual input. <s> BIB001 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> (7) <s> We propose a method for learning expressive energy-based policies for continuous states and actions, which has been feasible only in tabular domains before. We apply our method to learning maximum entropy policies, resulting into a new algorithm, called soft Q-learning, that expresses the optimal policy via a Boltzmann distribution. We use the recently proposed amortized Stein variational gradient descent to learn a stochastic sampling network that approximates samples from this distribution. The benefits of the proposed algorithm include improved exploration and compositionality that allows transferring skills between tasks, which we confirm in simulated experiments with swimming and walking robots. We also draw a connection to actor-critic methods, which can be viewed performing approximate inference on the corresponding energy-based model. <s> BIB002

In the case of Q-learning, the action is chosen according to the Q-function as in Eqn. 7. But DDPG chains the evaluation of Q after the action has already been chosen according to the policy. By correcting the Q-values towards the optimal values using the chosen action, we also update the policy towards the optimal action proposition. Thus two separate networks work at estimating Q * and π * . Asynchronous Advantage Actor Critic (A3C) BIB001 uses asynchronous gradient descent for optimization of deep neural network controllers. Deep reinforcement learning algorithms based on experience replay such as DQN and DDPG have demonstrated considerable success in difficult domains such as playing Atari games. However, experience replay uses a large amount of memory to store experience samples and requires off-policy learning algorithms. In A3C, instead of using an experience replay buffer, agents asynchronously execute on multiple parallel instances of the environment. In addition to the reducing correlation of the experiences, the parallel actor-learners have a stabilizing effect on training process. This simple setup enables a much larger spectrum of on-policy as well as off-policy reinforcement learning algorithms to be applied robustly using deep neural networks. A3C exceeded the performance of the previous state-of-theart at the time on the Atari domain while training for half the time on a single multi-core CPU instead of a GPU by combining several ideas. It also demonstrates how using an estimate of the value function as the previously explained baseline b reduces variance and improves convergence time. By defining the advantage as A π (a, s) = Q π (s, a) − V π (s), the expression of the policy gradient from Eqn. 5 is rewritten as The intuition of using advantage estimates rather than just discounted returns is to allow the agent to determine not just how good its actions were, but also how much better they turned out to be than expected, leading to reduced variance and more stable training. The A3C model also demonstrated good performance in 3D environments such as labyrinth exploration. Advantage Actor Critic (A2C) is a synchronous version of the asynchronous advantage actor critic model, that waits for each agent to finish its experience before conducting an update. The performance of both A2C and A3C is comparable. Most greedy policies must alternate between exploration and exploitation, and good exploration visits the states where the value estimate is uncertain. This way, exploration focuses on trying to find the most uncertain state paths as they bring valuable information. In addition to advantage, explained earlier, some methods use the entropy as the uncertainty quantity. Most A3C implementations include this as well. Two methods with common authors are energybased policies BIB002 and more recent and with widespread use, the Soft Actor Critic (SAC) algorithm , both rely on adding an entropy term to the reward function, so we update the policy objective from Eqn. 1 to Eqn. 8. We refer readers to for an in depth explanation of the expression shown here for illustration of how the entropy H is added.

Deep Reinforcement Learning for Autonomous Driving: A Survey <s> D. Model-based (vs. Model-free) & On/Off Policy methods <s> This paper extends previous work with Dyna, a class of architectures for intelligent systems based on approximating dynamic programming methods. Dyna architectures integrate trial-and-error (reinforcement) learning and execution-time planning into a single process operating alternately on the world and on a learned model of the world. In this paper, I present and show results for two Dyna architectures. The Dyna-PI architecture is based on dynamic programming's policy iteration method and can be related to existing AI ideas such as evaluation functions and universal plans (reactive systems). Using a navigation task, results are shown for a simple Dyna-PI system that simultaneously learns by trial and error, learns a world model, and plans optimal routes using the evolving world model. The Dyna-Q architecture is based on Watkins's Q-learning, a new kind of reinforcement learning. Dyna-Q uses a less familiar set of data structures than does Dyna-PI, but is arguably simpler to implement and use. We show that Dyna-Q architectures are easy to adapt for use in changing environments. <s> BIB001 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> D. Model-based (vs. Model-free) & On/Off Policy methods <s> Reinforcement learning, one of the most active research areas in artificial intelligence, is a computational approach to learning whereby an agent tries to maximize the total amount of reward it receives when interacting with a complex, uncertain environment. In Reinforcement Learning, Richard Sutton and Andrew Barto provide a clear and simple account of the key ideas and algorithms of reinforcement learning. Their discussion ranges from the history of the field's intellectual foundations to the most recent developments and applications. The only necessary mathematical background is familiarity with elementary concepts of probability. The book is divided into three parts. Part I defines the reinforcement learning problem in terms of Markov decision processes. Part II provides basic solution methods: dynamic programming, Monte Carlo methods, and temporal-difference learning. Part III presents a unified view of the solution methods and incorporates artificial neural networks, eligibility traces, and planning; the two final chapters present case studies and consider the future of reinforcement learning. <s> BIB002 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> D. Model-based (vs. Model-free) & On/Off Policy methods <s> We present a reinforcement learning architecture, Dyna-2, that encompasses both sample-based learning and sample-based search, and that generalises across states during both learning and search. We apply Dyna-2 to high performance Computer Go. In this domain the most successful planning methods are based on sample-based search algorithms, such as UCT, in which states are treated individually, and the most successful learning methods are based on temporal-difference learning algorithms, such as Sarsa, in which linear function approximation is used. In both cases, an estimate of the value function is formed, but in the first case it is transient, computed and then discarded after each move, whereas in the second case it is more permanent, slowly accumulating over many moves and games. The idea of Dyna-2 is for the transient planning memory and the permanent learning memory to remain separate, but for both to be based on linear function approximation and both to be updated by Sarsa. To apply Dyna-2 to 9x9 Computer Go, we use a million binary features in the function approximator, based on templates matching small fragments of the board. Using only the transient memory, Dyna-2 performed at least as well as UCT. Using both memories combined, it significantly outperformed UCT. Our program based on Dyna-2 achieved a higher rating on the Computer Go Online Server than any handcrafted or traditional search based program. <s> BIB003

In practical situations, interacting with the real environment could be limited due to many reasons including safety and cost. Learning a model for environment dynamics may reduce the amount of interactions required with the real environment. Moreover, exploration can be performed on the learned models. In the case of model-based approaches (e.g. Dyna-Q BIB001 , R-max ), agents attempt to learn the transition function T and reward function R, which can be used when making action selections. Keeping a model approximation of the environment means storing knowledge of its dynamics, and allows for fewer, and sometimes, costly environment interactions. By contrast, in model-free approaches such knowledge is not a requirement. Instead, model-free learners sample the underlying MDP directly in order to gain knowledge about the unknown model, in the form of value function estimates for example. In Dyna-2 BIB003 , the learning agent stores longterm and short-term memories, where a memory is defined as the set of features and corresponding parameters used by an agent to estimate the value function. Long-term memory is for general domain knowledge which is updated from real experience, while short-term memory is for specific local knowledge about the current situation, and the value function is a linear combination of long and short term memories. Learning algorithms can be on-policy or off-policy depending on whether the updates are conducted on fresh trajectories generated by the policy or by another policy, that could be generated by an older version of the policy or provided by an expert. On-policy methods such as SARSA , estimate the value of a policy while using the same policy for control. However, off-policy methods such as Q-learning , use two policies: the behavior policy, the policy used to generate behavior; and the target policy, the one being improved on. An advantage of this separation is that the target policy may be deterministic (greedy), while the behavior policy can continue to sample all possible actions, BIB002 .

Deep Reinforcement Learning for Autonomous Driving: A Survey <s> E. Deep reinforcement learning <s> TD-Gammon is a neural network that is able to teach itself to play backgammon solely by playing against itself and learning from the results, based on the TD(») reinforcement learning algorithm (Sutton 1988). Despite starting from random initial weights (and hence random initial strategy), TD-Gammon achieves a surprisingly strong level of play. With zero knowledge built in at the start of learning (i.e., given only a "raw" description of the board state), the network learns to play at a strong intermediate level. Furthermore, when a set of hand-crafted features is added to the network's input representation, the result is a truly staggering level of performance: the latest version of TD-Gammon is now estimated to play at a strong master level that is extremely close to the world's best human players. <s> BIB001 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> E. Deep reinforcement learning <s> The theory of reinforcement learning provides a normative account, deeply rooted in psychological and neuroscientific perspectives on animal behaviour, of how agents may optimize their control of an environment. To use reinforcement learning successfully in situations approaching real-world complexity, however, agents are confronted with a difficult task: they must derive efficient representations of the environment from high-dimensional sensory inputs, and use these to generalize past experience to new situations. Remarkably, humans and other animals seem to solve this problem through a harmonious combination of reinforcement learning and hierarchical sensory processing systems, the former evidenced by a wealth of neural data revealing notable parallels between the phasic signals emitted by dopaminergic neurons and temporal difference reinforcement learning algorithms. While reinforcement learning agents have achieved some successes in a variety of domains, their applicability has previously been limited to domains in which useful features can be handcrafted, or to domains with fully observed, low-dimensional state spaces. Here we use recent advances in training deep neural networks to develop a novel artificial agent, termed a deep Q-network, that can learn successful policies directly from high-dimensional sensory inputs using end-to-end reinforcement learning. We tested this agent on the challenging domain of classic Atari 2600 games. We demonstrate that the deep Q-network agent, receiving only the pixels and the game score as inputs, was able to surpass the performance of all previous algorithms and achieve a level comparable to that of a professional human games tester across a set of 49 games, using the same algorithm, network architecture and hyperparameters. This work bridges the divide between high-dimensional sensory inputs and actions, resulting in the first artificial agent that is capable of learning to excel at a diverse array of challenging tasks. <s> BIB002 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> E. Deep reinforcement learning <s> In this paper, we consider the task of learning control policies for text-based games. In these games, all interactions in the virtual world are through text and the underlying state is not observed. The resulting language barrier makes such environments challenging for automatic game players. We employ a deep reinforcement learning framework to jointly learn state representations and action policies using game rewards as feedback. This framework enables us to map text descriptions into vector representations that capture the semantics of the game states. We evaluate our approach on two game worlds, comparing against baselines using bag-ofwords and bag-of-bigrams for state representations. Our algorithm outperforms the baselines on both worlds demonstrating the importance of learning expressive representations. 1 <s> BIB003 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> E. Deep reinforcement learning <s> Experience replay lets online reinforcement learning agents remember and reuse experiences from the past. In prior work, experience transitions were uniformly sampled from a replay memory. However, this approach simply replays transitions at the same frequency that they were originally experienced, regardless of their significance. In this paper we develop a framework for prioritizing experience, so as to replay important transitions more frequently, and therefore learn more efficiently. We use prioritized experience replay in Deep Q-Networks (DQN), a reinforcement learning algorithm that achieved human-level performance across many Atari games. DQN with prioritized experience replay achieves a new state-of-the-art, outperforming DQN with uniform replay on 41 out of 49 games. <s> BIB004 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> E. Deep reinforcement learning <s> The popular Q-learning algorithm is known to overestimate action values under certain conditions. It was not previously known whether, in practice, such overestimations are common, whether they harm performance, and whether they can generally be prevented. In this paper, we answer all these questions affirmatively. In particular, we first show that the recent DQN algorithm, which combines Q-learning with a deep neural network, suffers from substantial overestimations in some games in the Atari 2600 domain. We then show that the idea behind the Double Q-learning algorithm, which was introduced in a tabular setting, can be generalized to work with large-scale function approximation. We propose a specific adaptation to the DQN algorithm and show that the resulting algorithm not only reduces the observed overestimations, as hypothesized, but that this also leads to much better performance on several games. <s> BIB005 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> E. Deep reinforcement learning <s> In recent years there have been many successes of using deep representations in reinforcement learning. Still, many of these applications use conventional architectures, such as convolutional networks, LSTMs, or auto-encoders. In this paper, we present a new neural network architecture for model-free reinforcement learning. Our dueling network represents two separate estimators: one for the state value function and one for the state-dependent action advantage function. The main benefit of this factoring is to generalize learning across actions without imposing any change to the underlying reinforcement learning algorithm. Our results show that this architecture leads to better policy evaluation in the presence of many similar-valued actions. Moreover, the dueling architecture enables our RL agent to outperform the state-of-the-art on the Atari 2600 domain. <s> BIB006 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> E. Deep reinforcement learning <s> Deep Reinforcement Learning has yielded proficient controllers for complex tasks. However, these controllers have limited memory and rely on being able to perceive the complete game screen at each decision point. To address these shortcomings, this article investigates the effects of adding recurrency to a Deep Q-Network (DQN) by replacing the first post-convolutional fully-connected layer with a recurrent LSTM. The resulting \textit{Deep Recurrent Q-Network} (DRQN), although capable of seeing only a single frame at each timestep, successfully integrates information through time and replicates DQN's performance on standard Atari games and partially observed equivalents featuring flickering game screens. Additionally, when trained with partial observations and evaluated with incrementally more complete observations, DRQN's performance scales as a function of observability. Conversely, when trained with full observations and evaluated with partial observations, DRQN's performance degrades less than DQN's. Thus, given the same length of history, recurrency is a viable alternative to stacking a history of frames in the DQN's input layer and while recurrency confers no systematic advantage when learning to play the game, the recurrent net can better adapt at evaluation time if the quality of observations changes. <s> BIB007 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> E. Deep reinforcement learning <s> The game of Go has long been viewed as the most challenging of classic games for artificial intelligence owing to its enormous search space and the difficulty of evaluating board positions and moves. Here we introduce a new approach to computer Go that uses ‘value networks’ to evaluate board positions and ‘policy networks’ to select moves. These deep neural networks are trained by a novel combination of supervised learning from human expert games, and reinforcement learning from games of self-play. Without any lookahead search, the neural networks play Go at the level of stateof-the-art Monte Carlo tree search programs that simulate thousands of random games of self-play. We also introduce a new search algorithm that combines Monte Carlo simulation with value and policy networks. Using this search algorithm, our program AlphaGo achieved a 99.8% winning rate against other Go programs, and defeated the human European Go champion by 5 games to 0. This is the first time that a computer program has defeated a human professional player in the full-sized game of Go, a feat previously thought to be at least a decade away. <s> BIB008

Tabular representations are the simplest way to store learned estimates (of e.g. values, policies or models), where each stateaction pair has a discrete estimate associated with it. When estimates are represented discretely, each additional feature tracked in the state leads to an exponential growth in the number of state-action pair values that must be stored . This problem is commonly referred to in the literature as the "curse of dimensionality", a term originally coined by Bellman . In simple environments this is rarely an issue, but it may lead to an intractable problem in real-world applications, due to memory and/or computational constraints. Learning over a large state-action space is possible, but may take an unacceptably long time to learn useful policies. Many realworld domains feature continuous state and/or action spaces; these can be discretised in many cases. However, large discretisation steps may limit the achievable performance in a domain, whereas small discretisation steps may result in a large stateaction space where obtaining a sufficient number of samples for each state-action pair is impractical. Alternatively, function approximation may be used to generalise across states and/or actions, whereby a function approximator is used to store and retrieve estimates. Function approximation is an active area of research in RL, offering a way to handle continuous state and/or action spaces, mitigate against the state-action space explosion and generalise prior experience to previously unseen state-action pairs. Tile coding is one of the simplest forms of function approximation, where one tile represents multiple states or state-action pairs . Neural networks are also commonly used to implement function approximation, one of the most famous examples being Tesuaro's application of RL to backgammon BIB001 . Recent work has applied deep neural networks as a function approximation method; this emerging paradigm is known as deep reinforcement learning (DRL). DRL algorithms have achieved human level performance (or above) on complex tasks such as playing Atari games BIB002 and playing the board game Go BIB008 . In DQN BIB002 it is demonstrated how a convolutional neural network can learn successful control policies from just raw video data for different Atari environments. The network was trained end-to-end and was not provided with any game specific information. The input to the convolutional neural network consists of an 84×84×4 where 4 consecutive frames are used to capture the temporal information. The first hidden layer consists of 32 filters of 8 × 8 with stride 4 and applies a rectifier non linearity. The second hidden layer consists of 64 filters of 4 × 4 with stride 2, followed by a rectifier nonlinearity. This is followed by a third convolutional layer that of 64 filters of 3 × 3 with stride1 followed by a rectifier. The final hidden layer is fully connected and consists of 512 rectifier units. The output layer is a fully-connected linear layer with a single output for each valid action. For DQN training stability, two networks are used while the parameters of the target network are fixed for a number of iterations while updating the online network parameters. For practical reasons, the Q(s, a) function is modeled as a deep neural network that predicts the value of all actions given the input state. Accordingly, deciding what action to take requires performing a single forward pass of the network. Moreover, in order to increase sample efficiency, experiences of the agent are stored in a replay memory (experience replay), where the Qlearning updates are conducted on randomly selected samples from the replay memory. This random selection breaks the correlation between successive samples. Experience replay enables reinforcement learning agents to remember and reuse experiences from the past where observed transitions are stored for some time, usually in a queue, and sampled uniformly from this memory to update the network. However, this approach simply replays transitions at the same frequency that they were originally experienced, regardless of their significance. An alternative method is to use two separate experience buckets, one for positive and one for negative rewards BIB003 . Then a fixed fraction from each bucket is selected to replay. This method is only applicable in domains that have a natural notion of binary experience. Experience replay has also been extended with a framework for prioritising experience BIB004 , where important transitions, based on the TD error, are replayed more frequently, leading to improved performance and faster training when compared to the standard experience replay approach. The max operator in standard Q-learning and DQN uses the same values both to select and to evaluate an action resulting in over optimistic value estimates. In Double DQN (D-DQN) BIB005 the over estimation problem in DQN is tackled where the greedy policy is evaluated according to the online network and uses the target network to estimate its value. It was shown that this algorithm not only yields more accurate value estimates, but leads to much higher scores on several games. In Dueling network architecture BIB006 the state value function and associated advantage function are estimated, and then combined together to estimate action value function. The advantage of the dueling architecture lies partly in its ability to learn the state-value function efficiently. In a single-stream architecture only the value for one of the actions is updated. However in dueling architecture, the value stream is updated with every update, allowing for better approximation of the state values, which in turn need to be accurate for temporal difference methods like Q-learning. DRQN BIB007 applied a modification to the DQN by combining a Long Short Term Memory (LSTM) with a Deep Q-Network. Accordingly, the DRQN is capable of integrating information across frames to detect information such as velocity of objects. DRQN showed to generalize its policies in case of complete observations and when trained on Atari games and evaluated against flickering games, it was shown that DRQN generalizes better than DQN.

Deep Reinforcement Learning for Autonomous Driving: A Survey <s> A. Reward shaping <s> We consider the El Farol bar problem, also known as the minority game (W. B. Arthur, The American Economic Review, 84 (1994) 406; D. Challet and Y. C. Zhang, Physica A, 256 (1998) 514). We view it as an instance of the general problem of how to configure the nodal elements of a distributed dynamical system so that they do not work at cross purposes, in that their collective dynamics avoids frustration and thereby achieves a provided global goal. We summarize a mathematical theory for such configuration applicable when (as in the bar problem) the global goal can be expressed as minimizing a global energy function and the nodes can be expressed as minimizers of local free energy functions. We show that a system designed with that theory performs nearly optimally for the bar problem. <s> BIB001 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> A. Reward shaping <s> Reward shaping has been proposed as a means to address the credit assignment problem in Multi-Agent Systems (MAS). Two popular shaping methods are Potential-Based Reward Shaping and difference rewards, and both have been shown to improve learning speed and the quality of joint policies learned by agents in single-objective MAS. In this work we discuss the theoretical implications of applying these approaches to multi-objective MAS, and evaluate their efficacy using a new multi-objective benchmark domain where the true set of Pareto optimal system utilities is known. <s> BIB002

As noted in Section III, the design of the reward function is crucial: RL agents seek to maximise the return from the reward function, therefore the optimal policy for a domain is defined with respect to the reward function. In many realworld application domains, learning may be difficult due to sparse and/or delayed rewards. RL agents typically learn how to act in their environment guided merely by the reward signal. Additional knowledge can be provided to a learner by the addition of a shaping reward to the reward naturally received from the environment, with the goal of improving learning speed and converged performance. This principle is referred to as reward shaping. The term shaping has its origins in the field of experimental psychology, and describes the idea of rewarding all behaviour that leads to the desired behaviour. Skinner discovered while training a rat to push a lever that any movement in the direction of the lever had to be rewarded to encourage the rat to complete the task. Analogously to the rat, a RL agent may take an unacceptably long time to discover its goal when learning from delayed rewards, and shaping offers an opportunity to speed up the learning process. Reward shaping allows a reward function to be engineered in a way to provide more frequent feedback signal on appropriate behaviours , which is especially useful in domains with sparse rewards. Generally, the return from the reward function is modified as follows: r = r+ f where r is the return from the original reward function R, f is the additional reward from a shaping function F, and r is the signal given to the agent by the augmented reward function R . Empirical evidence has shown that reward shaping can be a powerful tool to improve the learning speed of RL agents . However, it can have unintended consequences. The implication of adding a shaping reward is that a policy which is optimal for the augmented reward function R may not in fact also be optimal for the original reward function R. A classic example of reward shaping gone wrong for this exact reason is reported by where the experimented bicycle agent would turn in circle to stay upright rather than reach its goal. Difference rewards (D) BIB001 and potential-based reward shaping (PBRS) are two commonly used shaping approaches. Both D and PBRS have been successfully applied to a wide range of application domains and have the added benefit of convenient theoretical guarantees, meaning that they do not suffer from the same issues as the unprincipled reward shaping approaches described above (see e.g. - BIB002 ).

Deep Reinforcement Learning for Autonomous Driving: A Survey <s> C. Multi-objective reinforcement learning <s> Sequential decision-making problems with multiple objectives arise naturally in practice and pose unique challenges for research in decision-theoretic planning and learning, which has largely focused on single-objective settings. This article surveys algorithms designed for sequential decision-making problems with multiple objectives. Though there is a growing body of literature on this subject, little of it makes explicit under what circumstances special methods are needed to solve multi-objective problems. Therefore, we identify three distinct scenarios in which converting such a problem to a single-objective one is impossible, infeasible, or undesirable. Furthermore, we propose a taxonomy that classifies multi-objective methods according to the applicable scenario, the nature of the scalarization function (which projects multi-objective values to scalar ones), and the type of policies considered. We show how these factors determine the nature of an optimal solution, which can be a single policy, a convex hull, or a Pareto front. Using this taxonomy, we survey the literature on multi-objective methods for planning and learning. Finally, we discuss key applications of such methods and outline opportunities for future work. <s> BIB001 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> C. Multi-objective reinforcement learning <s> The majority of multi-agent system implementations aim to optimise agents’ policies with respect to a single objective, despite the fact that many real-world problem domains are inherently multi-objective in nature. Multi-objective multi-agent systems (MOMAS) explicitly consider the possible trade-offs between conflicting objective functions. We argue that, in MOMAS, such compromises should be analysed on the basis of the utility that these compromises have for the users of a system. As is standard in multi-objective optimisation, we model the user utility using utility functions that map value or return vectors to scalar values. This approach naturally leads to two different optimisation criteria: expected scalarised returns (ESR) and scalarised expected returns (SER). We develop a new taxonomy which classifies multi-objective multi-agent decision making settings, on the basis of the reward structures, and which and how utility functions are applied. This allows us to offer a structured view of the field, to clearly delineate the current state-of-the-art in multi-objective multi-agent decision making approaches and to identify promising directions for future research. Starting from the execution phase, in which the selected policies are applied and the utility for the users is attained, we analyse which solution concepts apply to the different settings in our taxonomy. Furthermore, we define and discuss these solution concepts under both ESR and SER optimisation criteria. We conclude with a summary of our main findings and a discussion of many promising future research directions in multi-objective multi-agent systems. <s> BIB002

In multi-objective reinforcement learning (MORL) the reward signal is a vector, where each component represents the performance on a different objective. The MORL framework was developed to handle sequential decision making problems where tradeoffs between conflicting objective functions must be considered. Examples of real-world problems with multiple objectives include selecting energy sources (tradeoffs between fuel cost and emissions) and watershed management (tradeoffs between generating electricity, preserving reservoir levels and supplying drinking water) . Solutions to MORL problems are often evaluated using the concept of Pareto dominance and MORL algorithms typically seek to learn or approximate the set of non-dominated solutions. MORL problems may be defined using the MDP or SG framework as appropriate, in a similar manner to single-objective problems. The main difference lies in the definition of the reward function: instead of returning a single scalar value r, the reward function R in multi-objective domains returns a vector r consisting of the rewards for each individual objective c ∈ C. Therefore, a regular MDP or SG can be extended to a Multi-Objective MDP (MOMDP) or Multi-Objective SG (MOSG) by modifying the return of the reward function. For a more complete overview of MORL beyond the brief summary presented in this section, the interested reader is referred to recent surveys BIB001 , BIB002 .

Deep Reinforcement Learning for Autonomous Driving: A Survey <s> D. State Representation Learning <s> This article reviews an emerging field that aims for autonomous reinforcement learning (RL) directly on sensor-observations. Straightforward end-to-end RL has recently shown remarkable success, but relies on large amounts of samples. As this is not feasible in robotics, we review two approaches to learn intermediate state representations from previous experiences: deep auto-encoders and slow-feature analysis. We analyze theoretical properties of the representations and point to potential improvements. <s> BIB001 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> D. State Representation Learning <s> Abstract Representation learning algorithms are designed to learn abstract features that characterize data. State representation learning (SRL) focuses on a particular kind of representation learning where learned features are in low dimension, evolve through time, and are influenced by actions of an agent. The representation is learned to capture the variation in the environment generated by the agent’s actions; this kind of representation is particularly suitable for robotics and control scenarios. In particular, the low dimension characteristic of the representation helps to overcome the curse of dimensionality, provides easier interpretation and utilization by humans and can help improve performance and speed in policy learning algorithms such as reinforcement learning. This survey aims at covering the state-of-the-art on state representation learning in the most recent years. It reviews different SRL methods that involve interaction with the environment, their implementations and their applications in robotics control tasks (simulated or real). In particular, it highlights how generic learning objectives are differently exploited in the reviewed algorithms. Finally, it discusses evaluation methods to assess the representation learned and summarizes current and future lines of research. <s> BIB002 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> D. State Representation Learning <s> Scaling end-to-end reinforcement learning to control real robots from vision presents a series of challenges, in particular in terms of sample efficiency. Against end-to-end learning, state representation learning can help learn a compact, efficient and relevant representation of states that speeds up policy learning, reducing the number of samples needed, and that is easier to interpret. We evaluate several state representation learning methods on goal based robotics tasks and propose a new unsupervised model that stacks representations and combines strengths of several of these approaches. This method encodes all the relevant features, performs on par or better than end-to-end learning, and is robust to hyper-parameters change. <s> BIB003

State Representation Learning (SRL) refers to feature extraction & dimensionality reduction to represent the state space with its history conditioned by the actions and environment of the agent. A complete review of SRL for control is discussed in BIB002 . In the simplest form SRL maps a high dimensional vector o t into a small dimensional latent space s t . The inverse operation decodes the state back into an estimate of the original observationô t . The agent then learns to map from the latent space to the action. Training the SRL chain is unsupervised in the sense that no labels are required. Reducing the dimension of the input effectively simplifies the task as it removes noise and decreases the domain's size as shown in BIB003 . SRL could be a simple auto-encoder (AE), though various methods exist for observation reconstruction such as Variational Auto-Encoders (VAE) or GANs, as well as forward models for predicting the next state or inverse models for predicting the action given a transition. A good learned state representation should be Markovian; i.e. it should encode all necessary information to be able to select an action based on the current state only, and not any previous states or actions BIB002 , BIB001 .

Deep Reinforcement Learning for Autonomous Driving: A Survey <s> E. Curriculum Learning & Hierarchical RL <s> For many types of machine learning algorithms, one can compute the statistically "optimal" way to select training data. In this paper, we review how optimal data selection techniques have been used with feedforward neural networks. We then show how the same principles may be used to select data for two alternative, statistically-based learning architectures: mixtures of Gaussians and locally weighted regression. While the techniques for neural networks are computationally expensive and approximate, the techniques for mixtures of Gaussians and locally weighted regression are both efficient and accurate. Empirically, we observe that the optimality criterion sharply decreases the number of training examples the learner needs in order to achieve good performance. <s> BIB001 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> E. Curriculum Learning & Hierarchical RL <s> Reinforcement learning is bedeviled by the curse of dimensionality: the number of parameters to be learned grows exponentially with the size of any compact encoding of a state. Recent attempts to combat the curse of dimensionality have turned to principled ways of exploiting temporal abstraction, where decisions are not required at each step, but rather invoke the execution of temporally-extended activities which follow their own policies until termination. This leads naturally to hierarchical control architectures and associated learning algorithms. We review several approaches to temporal abstraction and hierarchical organization that machine learning researchers have recently developed. Common to these approaches is a reliance on the theory of semi-Markov decision processes, which we emphasize in our review. We then discuss extensions of these ideas to concurrent activities, multiagent coordination, and hierarchical memory for addressing partial observability. Concluding remarks address open challenges facing the further development of reinforcement learning in a hierarchical setting. <s> BIB002 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> E. Curriculum Learning & Hierarchical RL <s> Humans and animals learn much better when the examples are not randomly presented but organized in a meaningful order which illustrates gradually more concepts, and gradually more complex ones. Here, we formalize such training strategies in the context of machine learning, and call them "curriculum learning". In the context of recent research studying the difficulty of training in the presence of non-convex training criteria (for deep deterministic and stochastic neural networks), we explore curriculum learning in various set-ups. The experiments show that significant improvements in generalization can be achieved. We hypothesize that curriculum learning has both an effect on the speed of convergence of the training process to a minimum and, in the case of non-convex criteria, on the quality of the local minima obtained: curriculum learning can be seen as a particular form of continuation method (a general strategy for global optimization of non-convex functions). <s> BIB003

Better learning performance can be achieved when the examples are organised in a meaningful order which illustrates more concepts gradually. Authors in BIB003 formalised training strategies in the context of machine learning that show that improvements in generalisation and in speed of convergence of the training. However, even humans do not agree about the order in which concepts should be introduced. Curriculum learning can be seen as a special form of transfer learning where the initial tasks are used to guide the agent to perform better on the target task. Curriculum learning is related to active learning BIB001 where the learning agent can focus on the interesting examples that stand near frontier of its knowledge and abilities. In such case, the pace at which the agent would move along a the curriculum is automated. Contrary to conventional or flat RL, HRL refers to the decomposition of complex agent behavior using temporal abstraction, such as the options framework BIB002 . Readers are directed towards a complete and compact review on advances in hierarchical reinforcement learning BIB002 .

Deep Reinforcement Learning for Autonomous Driving: A Survey <s> F. Learning from Demonstrations <s> Objective—To evaluate the pharmacokinetics of a novel commercial formulation of ivermectin after administration to goats. Animals—6 healthy adult goats. Procedure—Ivermectin (200 μg/kg) was initially administered IV to each goat, and plasma samples were obtained for 36 days. After a washout period of 3 weeks, each goat received a novel commercial formulation of ivermectin (200 μg/kg) by SC injection. Plasma samples were then obtained for 42 days. Drug concentrations were quantified by use of high-performance liquid chromatography with fluorescence detection. Results—Pharmacokinetics of ivermectin after IV administration were best described by a 2-compartment open model; values for main compartmental variables included volume of distribution at a steady state (9.94 L/kg), clearance (1.54 L/kg/d), and area under the plasma concentration-time curve (AUC; 143 [ng•d]/mL). Values for the noncompartmental variables included mean residence time (7.37 days), AUC (153 [ng•d]/mL), and clearance (1.43 L/kg/d). After ... <s> BIB001 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> F. Learning from Demonstrations <s> We consider learning in a Markov decision process where we are not explicitly given a reward function, but where instead we can observe an expert demonstrating the task that we want to learn to perform. This setting is useful in applications (such as the task of driving) where it may be difficult to write down an explicit reward function specifying exactly how different desiderata should be traded off. We think of the expert as trying to maximize a reward function that is expressible as a linear combination of known features, and give an algorithm for learning the task demonstrated by the expert. Our algorithm is based on using "inverse reinforcement learning" to try to recover the unknown reward function. We show that our algorithm terminates in a small number of iterations, and that even though we may never recover the expert's reward function, the policy output by the algorithm will attain performance close to that of the expert, where here performance is measured with respect to the expert's unknown reward function. <s> BIB002 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> F. Learning from Demonstrations <s> We consider reinforcement learning in systems with unknown dynamics. Algorithms such as E3 (Kearns and Singh, 2002) learn near-optimal policies by using "exploration policies" to drive the system towards poorly modeled states, so as to encourage exploration. But this makes these algorithms impractical for many systems; for example, on an autonomous helicopter, overly aggressive exploration may well result in a crash. In this paper, we consider the apprenticeship learning setting in which a teacher demonstration of the task is available. We show that, given the initial demonstration, no explicit exploration is necessary, and we can attain near-optimal performance (compared to the teacher) simply by repeatedly executing "exploitation policies" that try to maximize rewards. In finite-state MDPs, our algorithm scales polynomially in the number of states; in continuous-state linear dynamical systems, it scales polynomially in the dimension of the state. These results are proved using a martingale construction over relative losses. <s> BIB003 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> F. Learning from Demonstrations <s> We propose a new framework for estimating generative models via an adversarial process, in which we simultaneously train two models: a generative model G that captures the data distribution, and a discriminative model D that estimates the probability that a sample came from the training data rather than G. The training procedure for G is to maximize the probability of D making a mistake. This framework corresponds to a minimax two-player game. In the space of arbitrary functions G and D, a unique solution exists, with G recovering the training data distribution and D equal to ½ everywhere. In the case where G and D are defined by multilayer perceptrons, the entire system can be trained with backpropagation. There is no need for any Markov chains or unrolled approximate inference networks during either training or generation of samples. Experiments demonstrate the potential of the framework through qualitative and quantitative evaluation of the generated samples. <s> BIB004 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> F. Learning from Demonstrations <s> The game of Go has long been viewed as the most challenging of classic games for artificial intelligence owing to its enormous search space and the difficulty of evaluating board positions and moves. Here we introduce a new approach to computer Go that uses ‘value networks’ to evaluate board positions and ‘policy networks’ to select moves. These deep neural networks are trained by a novel combination of supervised learning from human expert games, and reinforcement learning from games of self-play. Without any lookahead search, the neural networks play Go at the level of stateof-the-art Monte Carlo tree search programs that simulate thousands of random games of self-play. We also introduce a new search algorithm that combines Monte Carlo simulation with value and policy networks. Using this search algorithm, our program AlphaGo achieved a 99.8% winning rate against other Go programs, and defeated the human European Go champion by 5 games to 0. This is the first time that a computer program has defeated a human professional player in the full-sized game of Go, a feat previously thought to be at least a decade away. <s> BIB005 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> F. Learning from Demonstrations <s> Consider learning a policy from example expert behavior, without interaction with the expert or access to reinforcement signal. One approach is to recover the expert's cost function with inverse reinforcement learning, then extract a policy from that cost function with reinforcement learning. This approach is indirect and can be slow. We propose a new general framework for directly extracting a policy from data, as if it were obtained by reinforcement learning following inverse reinforcement learning. We show that a certain instantiation of our framework draws an analogy between imitation learning and generative adversarial networks, from which we derive a model-free imitation learning algorithm that obtains significant performance gains over existing model-free methods in imitating complex behaviors in large, high-dimensional environments. <s> BIB006 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> F. Learning from Demonstrations <s> A long-standing goal of artificial intelligence is an algorithm that learns, tabula rasa, superhuman proficiency in challenging domains. Recently, AlphaGo became the first program to defeat a world champion in the game of Go. The tree search in AlphaGo evaluated positions and selected moves using deep neural networks. These neural networks were trained by supervised learning from human expert moves, and by reinforcement learning from self-play. Here we introduce an algorithm based solely on reinforcement learning, without human data, guidance or domain knowledge beyond game rules. AlphaGo becomes its own teacher: a neural network is trained to predict AlphaGo's own move selections and also the winner of AlphaGo's games. This neural network improves the strength of the tree search, resulting in higher quality move selection and stronger self-play in the next iteration. Starting tabula rasa, our new program AlphaGo Zero achieved superhuman performance, winning 100-0 against the previously published, champion-defeating AlphaGo. <s> BIB007 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> F. Learning from Demonstrations <s> Deep reinforcement learning (RL) has achieved several high profile successes in difficult decision-making problems. However, these algorithms typically require a huge amount of data before they reach reasonable performance. In fact, their performance during learning can be extremely poor. This may be acceptable for a simulator, but it severely limits the applicability of deep RL to many real-world tasks, where the agent must learn in the real environment. In this paper we study a setting where the agent may access data from previous control of the system. We present an algorithm, Deep Q-learning from Demonstrations (DQfD), that leverages small sets of demonstration data to massively accelerate the learning process even from relatively small amounts of demonstration data and is able to automatically assess the necessary ratio of demonstration data while learning thanks to a prioritized replay mechanism. DQfD works by combining temporal difference updates with supervised classification of the demonstrator's actions. We show that DQfD has better initial performance than Prioritized Dueling Double Deep Q-Networks (PDD DQN) as it starts with better scores on the first million steps on 41 of 42 games and on average it takes PDD DQN 83 million steps to catch up to DQfD's performance. DQfD learns to out-perform the best demonstration given in 14 of 42 games. In addition, DQfD leverages human demonstrations to achieve state-of-the-art results for 11 games. Finally, we show that DQfD performs better than three related algorithms for incorporating demonstration data into DQN. <s> BIB008 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> F. Learning from Demonstrations <s> Generative Adversarial Networks (GAN) have gained a lot of popularity from their introduction in 2014 till present. Research on GAN is rapidly growing and there are many variants of the original GAN focusing on various aspects of deep learning. GAN are perceived as the most impactful direction of machine learning in the last decade. This paper focuses on the application of GAN in autonomous driving including topics such as advanced data augmentation, loss function learning, semi-supervised learning, etc. We formalize and review key applications of adversarial techniques and discuss challenges and open problems to be addressed. <s> BIB009

Learning from Demonstrations (LfD) is used by humans to acquire new skills in an expert to learner knowledge transmission process. LfD is important for initial exploration where reward signals are too sparse or the input domain is too large to cover. In LfD, an agent learns to perform a task from demonstrations, usually in the form of state-action pairs, provided by an expert without any feedback rewards. However, high quality and diverse demonstrations are hard to collect, leading to learning sub-optimal policies. Accordingly, learning merely from demonstrations can be used to initialize the learning agent with a good or safe policy, and then reinforcement learning can be conducted to enable the discovery of a better policy by interacting with the environment. Combining demonstrations and reinforcement learning has been conducted in recent research. AlphaGo BIB005 , combines search tree with deep neural networks, initializes the policy network by supervised learning on state-action pairs provided by recorded games played by human experts. Additionally, a value network is trained to tell how desirable a board state is. By conducting self-play and reinforcement learning, AlphaGo is able to discover new stronger actions and learn from its mistakes, achieving super human performance. More recently, AlphaZero BIB007 , developed by the same team, proposed a general framework for self-play models. AlphaZero is trained entirely using reinforcement learning and self play, starting from completely random play, and requires no prior knowledge of human players. AlphaZero taught itself from scratch how to master the games of chess, shogi, and Go game, beating a world-champion program in each case. In BIB003 it is shown that given the initial demonstration, no explicit exploration is necessary, and we can attain near-optimal performance. Measuring the divergence between the current policy and the expert policy for optimization is proposed in . DQfD BIB008 pre-trains the agent and uses expert demonstrations by adding them into the replay buffer with additional priority. Moreover, a training framework that combines learning from both demonstrations and reinforcement learning is proposed in for fast learning agents. Two policies close to maximizing the reward function can still have large differences in behaviour. To avoid degenerating a solution which would fit the reward but not the original behaviour, authors BIB002 proposed a method for enforcing that the optimal policy learnt over the rewards should still match the observed policy in behavior. Behavior Cloning (BC) is applied as a supervised learning that maps states to actions based on demonstrations provided by an expert. On the other hand, Inverse Reinforcement Learning (IRL) is about inferring the reward function that justifies demonstrations of the expert. IRL is the problem of extracting a reward function given observed, optimal behavior BIB001 . A key motivation is that the reward function provides a succinct and robust definition of a task. Generally, IRL algorithms can be expensive to run, requiring reinforcement learning in an inner loop between cost estimation to policy training and evaluation. Generative Adversarial Imitation Learning (GAIL) BIB006 introduces a way to avoid this expensive inner loop. In practice, GAIL trains a policy close enough to the expert policy to fool a discriminator. This process is similar to Generative Adversarial Networks (GANs) BIB004 , BIB009 . The resulting policy must travel the same MDP states as the expert, or the discriminator would pick up the differences. The theory behind GAIL is an equation simplification: qualitatively, if IRL is going from demonstrations to a cost function and RL from a cost function to a policy, then we should altogether be able to go from demonstration to policy in a single equation while avoiding the cost function estimation.

Deep Reinforcement Learning for Autonomous Driving: A Survey <s> A. State Spaces, Action Spaces and Rewards <s> Learning, planning, and representing knowledge at multiple levels of temporal ab- straction are key, longstanding challenges for AI. In this paper we consider how these challenges can be addressed within the mathematical framework of reinforce- ment learning and Markov decision processes (MDPs). We extend the usual notion of action in this framework to include options—closed-loop policies for taking ac- tion over a period of time. Examples of options include picking up an object, going to lunch, and traveling to a distant city, as well as primitive actions such as mus- cle twitches and joint torques. Overall, we show that options enable temporally abstract knowledge and action to be included in the reinforcement learning frame- work in a natural and general way. In particular, we show that options may be used interchangeably with primitive actions in planning methods such as dynamic pro- gramming and in learning methods such as Q-learning. Formally, a set of options defined over an MDP constitutes a semi-Markov decision process (SMDP), and the theory of SMDPs provides the foundation for the theory of options. However, the most interesting issues concern the interplay between the underlying MDP and the SMDP and are thus beyond SMDP theory. We present results for three such cases: 1) we show that the results of planning with options can be used during execution to interrupt options and thereby perform even better than planned, 2) we introduce new intra-option methods that are able to learn about an option from fragments of its execution, and 3) we propose a notion of subgoal that can be used to improve the options themselves. All of these results have precursors in the existing literature; the contribution of this paper is to establish them in a simpler and more general setting with fewer changes to the existing reinforcement learning framework. In particular, we show that these results can be obtained without committing to (or ruling out) any particular approach to state abstraction, hierarchy, function approximation, or the macro-utility problem. <s> BIB001 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> A. State Spaces, Action Spaces and Rewards <s> Robust perception-action models should be learned from training data with diverse visual appearances and realistic behaviors, yet current approaches to deep visuomotor policy learning have been generally limited to in-situ models learned from a single vehicle or simulation environment. We advocate learning a generic vehicle motion model from large scale crowd-sourced video data, and develop an end-to-end trainable architecture for learning to predict a distribution over future vehicle egomotion from instantaneous monocular camera observations and previous vehicle state. Our model incorporates a novel FCN-LSTM architecture, which can be learned from large-scale crowd-sourced vehicle action data, and leverages available scene segmentation side tasks to improve performance under a privileged learning paradigm. We provide a novel large-scale dataset of crowd-sourced driving behavior suitable for training our model, and report results predicting the driver action on held out sequences across diverse conditions. <s> BIB002 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> A. State Spaces, Action Spaces and Rewards <s> We introduce CARLA, an open-source simulator for autonomous driving research. CARLA has been developed from the ground up to support development, training, and validation of autonomous urban driving systems. In addition to open-source code and protocols, CARLA provides open digital assets (urban layouts, buildings, vehicles) that were created for this purpose and can be used freely. The simulation platform supports flexible specification of sensor suites and environmental conditions. We use CARLA to study the performance of three approaches to autonomous driving: a classic modular pipeline, an end-to-end model trained via imitation learning, and an end-to-end model trained via reinforcement learning. The approaches are evaluated in controlled scenarios of increasing difficulty, and their performance is examined via metrics provided by CARLA, illustrating the platform's utility for autonomous driving research. The supplementary video can be viewed at this https URL <s> BIB003 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> A. State Spaces, Action Spaces and Rewards <s> Autonomous driving is a challenging domain that entails multiple aspects: a vehicle should be able to drive to its destination as fast as possible while avoiding collision, obeying traffic rules and ensuring the comfort of passengers. In this paper, we present a deep learning variant of thresholded lexicographic Q-learning for the task of urban driving. Our multi-objective DQN agent learns to drive on multi-lane roads and intersections, yielding and changing lanes according to traffic rules. We also propose an extension for factored Markov Decision Processes to the DQN architecture that provides auxiliary features for the Q function. This is shown to significantly improve data efficiency. \footnoteData efficiency as measured by the number of training steps required to achieve similar performance. We then show that the learned policy is able to zero-shot transfer to a ring road without sacrificing performance. <s> BIB004 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> A. State Spaces, Action Spaces and Rewards <s> Reinforcement learning has steadily improved and outperform human in lots of traditional games since the resurgence of deep neural network. However, these success is not easy to be copied to autonomous driving because the state spaces in real world are extreme complex and action spaces are continuous and fine control is required. Moreover, the autonomous driving vehicles must also keep functional safety under the complex environments. To deal with these challenges, we first adopt the deep deterministic policy gradient (DDPG) algorithm, which has the capacity to handle complex state and action spaces in continuous domain. We then choose The Open Racing Car Simulator (TORCS) as our environment to avoid physical damage. Meanwhile, we select a set of appropriate sensor information from TORCS and design our own rewarder. In order to fit DDPG algorithm to TORCS, we design our network architecture for both actor and critic inside DDPG paradigm. To demonstrate the effectiveness of our model, We evaluate on different modes in TORCS and show both quantitative and qualitative results. <s> BIB005 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> A. State Spaces, Action Spaces and Rewards <s> Urban autonomous driving decision making is challenging due to complex road geometry and multi-agent interactions. Current decision making methods are mostly manually designing the driving policy, which might result in suboptimal solutions and is expensive to develop, generalize and maintain at scale. On the other hand, with reinforcement learning (RL), a policy can be learned and improved automatically without any manual designs. However, current RL methods generally do not work well on complex urban scenarios. In this paper, we propose a framework to enable model-free deep reinforcement learning in challenging urban autonomous driving scenarios. We design a specific input representation and use visual encoding to capture the low-dimensional latent states. Several state-of-the-art model-free deep RL algorithms are implemented into our framework, with several tricks to improve their performance. We evaluate our method in a challenging roundabout task with dense surrounding vehicles in a high-definition driving simulator. The result shows that our method can solve the task well and is significantly better than the baseline. <s> BIB006

To successfully apply DRL to autonomous driving tasks, designing appropriate state spaces, action spaces, and reward functions is important. Leurent et al. provided a comprehensive review of the different state and action representations which are used in autonomous driving research. Commonly used state space features for an autonomous vehicle include: position, heading and velocity of ego-vehicle, as well as other obstacles in the sensor view extent of the ego-vehicle. To avoid variations in the dimension of the state space, a Cartesian or Polar occupancy grid around the ego vehicle is frequently employed. This is further augmented with lane information such as lane number (ego-lane or others), path curvature, future trajectory of the ego-vehicle, longitudinal information such as Time-to-collision (TTC), and finally scene information such as traffic laws and signal locations. Using raw sensor data such as camera images, LiDAR, radar, etc. provides the benefit of finer contextual information, while using condensed abstracted data reduces the complexity of the state space. In between, a mid-level representation such as 2D bird eye view (BEV) is sensor agnostic but still close to the spatial organization of the scene. Fig. 3 is an illustration of a top down view showing an occupancy grid, past and projected trajectories, and semantic information about the scene such as the position of traffic lights. This intermediary format retains the spatial layout of roads when graph-based representations would not. Some simulators offer this view such as Carla or Flow (see Table V -C). A vehicle policy must control a number of different actuators. Continuous-valued actuators for vehicle control include steering angle, throttle and brake. Other actuators such as gear changes are discrete. To reduce complexity and allow the application of DRL algorithms which work with discrete action spaces only (e.g. DQN), an action space may be discretised uniformly by dividing the range of continuous actuators such as steering angle, throttle and brake into equal-sized bins. Discretisation in log-space has also been suggested, as many steering angles which are selected in practice are close to the centre BIB002 . Discretisation does have disadvantages however; it can lead to jerky or unstable trajectories if the step values between actions are too large. Furthermore, when selecting the number of bins for an actuator there is a tradeoff between having enough discrete steps to allow for smooth control, and not having so many steps that action selections become prohibitively expensive to evaluate. As an alternative to discretisation, continuous values for actuators may also be handled by DRL algorithms which learn a policy directly, (e.g. DDPG). Temporal abstractions options framework BIB001 ) may also be employed to simplify the process of selecting actions, where agents select options instead of low-level actions. These options represent a subpolicy that could extend a primitive action over multiple time steps. How to design reward functions to train DRL agents for autonomous driving is still very much an open question. Examples of criteria which have been integrated into reward functions when training DRL agents for AD tasks include: distance travelled towards a destination BIB003 , speed of the ego vehicle BIB003 - , keeping the ego vehicle at a standstill BIB006 , collisions with other road users or scene objects BIB003 , BIB004 , BIB006 , infractions on sidewalks BIB003 , keeping in lane BIB003 , BIB005 - , maintaining comfort and stability (avoiding extreme acceleration, braking or steering) BIB004 , BIB006 , , and following traffic rules BIB004 . Many applications of DRL for AD use a combination of more than one criterion in the reward function; a weighted sum is often used to linearly scalarise these components into a single reward term to be used during learning (see e.g. BIB003 ). As AD is essentially a multi-objective problem, methods from the field multi-objective RL such as thresholded lexicographic ordering may be easily applied and have been demonstrated to work well (see e.g. BIB004 ).

Deep Reinforcement Learning for Autonomous Driving: A Survey <s> B. Motion Planning & Trajectory optimization <s> Providing an efficient strategy to navigate safely through unsignaled intersections is a difficult task that requires determining the intent of other drivers. We explore the effectiveness of Deep Reinforcement Learning to handle intersection problems. Using recent advances in Deep RL, we are able to learn policies that surpass the performance of a commonly-used heuristic approach in several metrics including task completion time and goal success rate and have limited ability to generalize. We then explore a system's ability to learn active sensing behaviors to enable navigating safely in the case of occlusions. Our analysis, provides insight into the intersection handling problem, the solutions learned by the network point out several shortcomings of current rule-based methods, and the failures of our current deep reinforcement learning system point to future research directions. <s> BIB001 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> B. Motion Planning & Trajectory optimization <s> Behavior-related research areas such as motion prediction/planning, representation/imitation learning, behavior modeling/generation, and algorithm testing, require support from high-quality motion datasets containing interactive driving scenarios with different driving cultures. In this paper, we present an INTERnational, Adversarial and Cooperative moTION dataset (INTERACTION dataset) in interactive driving scenarios with semantic maps. Five features of the dataset are highlighted. 1) The interactive driving scenarios are diverse, including urban/highway/ramp merging and lane changes, roundabouts with yield/stop signs, signalized intersections, intersections with one/two/all-way stops, etc. 2) Motion data from different countries and different continents are collected so that driving preferences and styles in different cultures are naturally included. 3) The driving behavior is highly interactive and complex with adversarial and cooperative motions of various traffic participants. Highly complex behavior such as negotiations, aggressive/irrational decisions and traffic rule violations are densely contained in the dataset, while regular behavior can also be found from cautious car-following, stop, left/right/U-turn to rational lane-change and cycling and pedestrian crossing, etc. 4) The levels of criticality span wide, from regular safe operations to dangerous, near-collision maneuvers. Real collision, although relatively slight, is also included. 5) Maps with complete semantic information are provided with physical layers, reference lines, lanelet connections and traffic rules. The data is recorded from drones and traffic cameras. Statistics of the dataset in terms of number of entities and interaction density are also provided, along with some utilization examples in a variety of behavior-related research areas. The dataset can be downloaded via this https URL. <s> BIB002

Motion planning is the task of ensuring the existence of a path between target and destination points. This is necessary to plan trajectories for vehicles over prior maps usually augmented with semantic information. Path planning in dynamic environments and varying vehicle dynamics is a key problem in autonomous driving, for example negotiating right to pass through in an intersection BIB001 , merging into highways. Recent work by authors BIB002 contains real world motions by various traffic actors, observed in diverse interactive driving scenarios. Recently, authors demonstrated an application of DRL (DDPG) for AD using a full-sized autonomous vehicle

Deep Reinforcement Learning for Autonomous Driving: A Survey <s> Motion Planning <s> In this paper, we present a learning method to solve the vehicle overtaking problem, which demands a multitude of abilities from the agent to tackle multiple criteria. To handle this problem, we propose to adopt a multiple-goal reinforcement learning (MGRL) framework as the basis of our solution. By considering seven different goals, either Q-learning (QL) or double-action QL is employed to determine action decisions based on whether the other vehicles interact with the agent for that particular goal. Furthermore, a fusion function is proposed according to the importance of each goal before arriving to an overall but consistent action decision. This offers a powerful approach for dealing with demanding situations such as overtaking, particularly when a number of other vehicles are within the proximity of the agent and are traveling at different and varying speeds. A large number of overtaking cases have been simulated to demonstrate its effectiveness. From the results, it can be concluded that the proposed method is capable of the following: 1) making correct action decisions for overtaking; 2) avoiding collisions with other vehicles; 3) reaching the target at reasonable time; 4) keeping almost steady speed; and 5) maintaining almost steady heading angle. In addition, it should also be noted that the proposed method performs lane keeping well when not overtaking and lane changing effectively when overtaking is in progress. <s> BIB001 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> Motion Planning <s> Multiple automakers have in development or in production automated driving systems (ADS) that offer freeway-pilot functions. This type of ADS is typically limited to restricted-access freeways only, that is, the transition from manual to automated modes takes place only after the ramp merging process is completed manually. One major challenge to extend the automation to ramp merging is that the automated vehicle needs to incorporate and optimize long-term objectives (e.g. successful and smooth merge) when near-term actions must be safely executed. Moreover, the merging process involves interactions with other vehicles whose behaviors are sometimes hard to predict but may influence the merging vehicle optimal actions. To tackle such a complicated control problem, we propose to apply Deep Reinforcement Learning (DRL) techniques for finding an optimal driving policy by maximizing the long-term reward in an interactive environment. Specifically, we apply a Long Short-Term Memory (LSTM) architecture to model the interactive environment, from which an internal state containing historical driving information is conveyed to a Deep Q-Network (DQN). The DQN is used to approximate the Q-function, which takes the internal state as input and generates Q-values as output for action selection. With this DRL architecture, the historical impact of interactive environment on the long-term reward can be captured and taken into account for deciding the optimal control policy. The proposed architecture has the potential to be extended and applied to other autonomous driving scenarios such as driving through a complex intersection or changing lanes under varying traffic flow conditions. <s> BIB002 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> Motion Planning <s> A* is a popular path-finding algorithm, but it can only be applied to those domains where a good heuristic function is known. Inspired by recent methods combining Deep Neural Networks (DNNs) and trees, this study demonstrates how to train a heuristic represented by a DNN and combine it with A*. This new algorithm which we call aleph-star can be used efficiently in domains where the input to the heuristic could be processed by a neural network. We compare aleph-star to N-Step Deep Q-Learning (DQN Mnih et al. 2013) in a driving simulation with pixel-based input, and demonstrate significantly better performance in this scenario. <s> BIB003 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> Motion Planning <s> Providing an efficient strategy to navigate safely through unsignaled intersections is a difficult task that requires determining the intent of other drivers. We explore the effectiveness of Deep Reinforcement Learning to handle intersection problems. Using recent advances in Deep RL, we are able to learn policies that surpass the performance of a commonly-used heuristic approach in several metrics including task completion time and goal success rate and have limited ability to generalize. We then explore a system's ability to learn active sensing behaviors to enable navigating safely in the case of occlusions. Our analysis, provides insight into the intersection handling problem, the solutions learned by the network point out several shortcomings of current rule-based methods, and the failures of our current deep reinforcement learning system point to future research directions. <s> BIB004

Learn to plan trajectories dynamically and optimize cost function to provide smooth control behavior of vehicle. Inverse RL is utilized to learn optimal reward function (or shaping) from experts. Authors propose to learn a heuristic function for the A * algorithm using a DQN over image-based input obstacle map BIB003 Overtaking Authors BIB001 propose Multi-goal RL (MGRL) framework to learn overtaking policy while avoiding collisions & maintain steady speed. Intersections/Merging Ego-vehicle required to negotiate intersections and merges into highways BIB004 , Ramp merging is tackled in BIB002 , where DRL is applied to find an optimal driving policy using LSTM for producing an internal state containing historical driving information and DQN for Q-function approximation.

Deep Reinforcement Learning for Autonomous Driving: A Survey <s> Lane Keep <s> Reinforcement learning is considered to be a strong AI paradigm which can be used to teach machines through interaction with the environment and learning from their mistakes, but it has not yet been successfully used for automotive applications. There has recently been a revival of interest in the topic, however, driven by the ability of deep learning algorithms to learn good representations of the environment. Motivated by Google DeepMind's successful demonstrations of learning for games from Breakout to Go, we will propose different methods for autonomous driving using deep reinforcement learning. This is of particular interest as it is difficult to pose autonomous driving as a supervised learning problem as it has a strong interaction with the environment including other vehicles, pedestrians and roadworks. As this is a relatively new area of research for autonomous driving, we will formulate two main categories of algorithms: 1) Discrete actions category, and 2) Continuous actions category. For the discrete actions category, we will deal with Deep Q-Network Algorithm (DQN) while for the continuous actions category, we will deal with Deep Deterministic Actor Critic Algorithm (DDAC). In addition to that, We will also discover the performance of these two categories on an open source car simulator for Racing called (TORCS) which stands for The Open Racing car Simulator. Our simulation results demonstrate learning of autonomous maneuvering in a scenario of complex road curvatures and simple interaction with other vehicles. Finally, we explain the effect of some restricted conditions, put on the car during the learning phase, on the convergence time for finishing its learning phase. <s> BIB001 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> Lane Keep <s> Reinforcement learning is considered to be a strong AI paradigm which can be used to teach machines through interaction with the environment and learning from their mistakes. Despite its perceived utility, it has not yet been successfully applied in automotive applications. Motivated by the successful demonstrations of learning of Atari games and Go by Google DeepMind, we propose a framework for autonomous driving using deep reinforcement learning. This is of particular relevance as it is difficult to pose autonomous driving as a supervised learning problem due to strong interactions with the environment including other vehicles, pedestrians and roadworks. As it is a relatively new area of research for autonomous driving, we provide a short overview of deep reinforcement learning and then describe our proposed framework. It incorporates Recurrent Neural Networks for information integration, enabling the car to handle partially observable scenarios. It also integrates the recent work on attention models to focus on relevant information, thereby reducing the computational complexity for deployment on embedded hardware. The framework was tested in an open source 3D car racing simulator called TORCS. Our simulation results demonstrate learning of autonomous maneuvering in a scenario of complex road curvatures and simple interaction of other vehicles. <s> BIB002

Ego-vehicle follows the lane. Authors BIB001 propose a DRL system for discrete actions (DQN) and continuous actions (DDAC) using the TORCS simulator (see Table V -C), study concludes that continuous actions provide smoother trajectories, while more restricted termination conditions lead to the slower convergence time to learn. Automated parking Learn policies to automatically park the vehicle BIB002 .

Deep Reinforcement Learning for Autonomous Driving: A Survey <s> TABLE I LIST OF AD TASKS THAT REQUIRE D(RL) TO LEARN A POLICY OR BEHAVIOR. <s> We introduce Embed to Control (E2C), a method for model learning and control of non-linear dynamical systems from raw pixel images. E2C consists of a deep generative model, belonging to the family of variational autoencoders, that learns to generate image trajectories from a latent space in which the dynamics is constrained to be locally linear. Our model is derived directly from an optimal control formulation in latent space, supports long-term prediction of image sequences and exhibits strong performance on a variety of complex control problems. <s> BIB001 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> TABLE I LIST OF AD TASKS THAT REQUIRE D(RL) TO LEARN A POLICY OR BEHAVIOR. <s> Data-efficient learning in continuous state-action spaces using very high-dimensional observations remains a key challenge in developing fully autonomous systems. In this paper, we consider one instance of this challenge, the pixels-to-torques problem, where an agent must learn a closed-loop control policy from pixel information only. We introduce a data-efficient, model-based reinforcement learning algorithm that learns such a closed-loop policy directly from pixel information. The key ingredient is a deep dynamical model that uses deep auto-encoders to learn a low-dimensional embedding of images jointly with a prediction model in this low-dimensional feature space. This joint learning ensures that not only static properties of the data are accounted for, but also dynamic properties. This is crucial for long-term predictions, which lie at the core of the adaptive model predictive control strategy that we use for closed-loop control. Compared to state-of-the-art reinforcement learning methods, our approach learns quickly, scales to high-dimensional state spaces and facilitates fully autonomous learning from pixels to torques. <s> BIB002 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> TABLE I LIST OF AD TASKS THAT REQUIRE D(RL) TO LEARN A POLICY OR BEHAVIOR. <s> Models that can simulate how environments change in response to actions can be used by agents to plan and act efficiently. We improve on previous environment simulators from high-dimensional pixel observations by introducing recurrent neural networks that are able to make temporally and spatially coherent predictions for hundreds of time-steps into the future. We present an in-depth analysis of the factors affecting performance, providing the most extensive attempt to advance the understanding of the properties of these models. We address the issue of computationally inefficiency with a model that does not need to generate a high-dimensional image at each time-step. We show that our approach can be used to improve exploration and is adaptable to many diverse environments, namely 10 Atari games, a 3D car racing environment, and complex 3D mazes. <s> BIB003 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> TABLE I LIST OF AD TASKS THAT REQUIRE D(RL) TO LEARN A POLICY OR BEHAVIOR. <s> This manuscript surveys reinforcement learning from the perspective of optimization and control with a focus on continuous control applications. It surveys the general formulation, terminology, and typical experimental implementations of reinforcement learning and reviews competing solution paradigms. In order to compare the relative merits of various techniques, this survey presents a case study of the Linear Quadratic Regulator (LQR) with unknown dynamics, perhaps the simplest and best-studied problem in optimal control. The manuscript describes how merging techniques from learning theory and control can provide non-asymptotic characterizations of LQR performance and shows that these characterizations tend to match experimental behavior. In turn, when revisiting more complex applications, many of the observed phenomena in LQR persist. In particular, theory and experiment demonstrate the role and importance of models and the cost of generality in reinforcement learning algorithms. This survey concludes with a discussion of some of the challenges in designing learning systems that safely and reliably interact with complex and uncertain environments and how tools from reinforcement learning and control might be combined to approach these challenges. <s> BIB004 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> TABLE I LIST OF AD TASKS THAT REQUIRE D(RL) TO LEARN A POLICY OR BEHAVIOR. <s> A common belief in model-free reinforcement learning is that methods based on random search in the parameter space of policies exhibit significantly worse sample complexity than those that explore the space of actions. We dispel such beliefs by introducing a random search method for training static, linear policies for continuous control problems, matching state-of-the-art sample efficiency on the benchmark MuJoCo locomotion tasks. Our method also finds a nearly optimal controller for a challenging instance of the Linear Quadratic Regulator, a classical problem in control theory, when the dynamics are not known. Computationally, our random search algorithm is at least 15 times more efficient than the fastest competing model-free methods on these benchmarks. We take advantage of this computational efficiency to evaluate the performance of our method over hundreds of random seeds and many different hyperparameter configurations for each benchmark task. Our simulations highlight a high variability in performance in these benchmark tasks, suggesting that commonly used estimations of sample efficiency do not adequately evaluate the performance of RL algorithms. <s> BIB005

[92]. The system was first trained in simulation, before being trained in real time using on board computers, and was able to learn to follow a lane, successfully completing a realworld trial on a 250 metre section of road. Model-based deep RL algorithms have been proposed for learning models and policies directly from raw pixel inputs BIB001 , BIB002 . In BIB003 , deep neural networks have been used to generate predictions in simulated environments over hundreds of time steps. RL is also suitable for Control. Classical optimal control methods like LQR/iLQR are compared with RL methods in BIB004 . Classical RL methods are used to perform optimal control in stochastic settings, for example the Linear Quadratic Regulator (LQR) in linear regimes and iterative LQR (iLQR) for non-linear regimes are utilized. A recent study in BIB005 demonstrates that random search over the parameters for a policy network can perform as well as LQR.

Deep Reinforcement Learning for Autonomous Driving: A Survey <s> C. Simulator & Scenario generation tools <s> We present a framework for reinforcement learning (RL) in a scenario where multiple simulators are available with decreasing amounts of fidelity to the real-world learning scenario. Our framework is designed to limit the number of samples used in each successively higher-fidelity/cost simulator by allowing the agent to choose to run trajectories at the lowest level that will still provide it with information. The approach transfers state-action Q-values from lower-fidelity models as heuristics for the “Knows What It Knows” family of RL algorithms, which is applicable over a wide range of possible dynamics and reward representations. Theoretical proofs of the framework's sample complexity are given and empirical results are demonstrated on a remote controlled car with multiple simulators. The approach allows RL algorithms to find near-optimal policies for the real world with fewer expensive real-world samples than previous transfer approaches or learning without simulators. <s> BIB001 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> C. Simulator & Scenario generation tools <s> A multitude of publicly-available driving datasets and data platforms have been raised for autonomous vehicles (AV). However, the heterogeneities of databases in size, structure and driving context make existing datasets practically ineffective due to a lack of uniform frameworks and searchable indexes. In order to overcome these limitations on existing public datasets, this paper proposes a data unification framework based on traffic primitives with ability to automatically unify and label heterogeneous traffic data. This is achieved by two steps: 1) Carefully arrange raw multidimensional time series driving data into a relational database and then 2) automatically extract labeled and indexed traffic primitives from traffic data through a Bayesian nonparametric learning method. Finally, we evaluate the effectiveness of our developed framework using the collected real vehicle data. <s> BIB002 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> C. Simulator & Scenario generation tools <s> This paper presents a systematic review of the perception systems and simulators for autonomous vehicles (AV). This work has been divided into three parts. In the first part, perception systems are categorized as environment perception systems and positioning estimation systems. The paper presents the physical fundamentals, principle functioning, and electromagnetic spectrum used to operate the most common sensors used in perception systems (ultrasonic, RADAR, LiDAR, cameras, IMU, GNSS, RTK, etc.). Furthermore, their strengths and weaknesses are shown, and the quantification of their features using spider charts will allow proper selection of different sensors depending on 11 features. In the second part, the main elements to be taken into account in the simulation of a perception system of an AV are presented. For this purpose, the paper describes simulators for model-based development, the main game engines that can be used for simulation, simulators from the robotics field, and lastly simulators used specifically for AV. Finally, the current state of regulations that are being applied in different countries around the world on issues concerning the implementation of autonomous vehicles is presented. <s> BIB003

Autonomous driving datasets address supervised learning setup with training sets containing image, label pairs for various modalities. Reinforcement learning requires an environment where state-action pairs can be recovered while modelling dynamics of the vehicle state, environment as well as the stochasticity in the movement and actions of the environment and agent respectively. Various simulators are actively used for training and validating reinforcement learning algorithms. Table V-C summarises various high fidelity perception simulators capable of simulating cameras, LiDARs and radar. Some simulators are also capable of providing the vehicle state and dynamics. A complete review of sensors and simulators utilised within the autonomous driving community is available in BIB003 for readers. Learned driving policies are stress tested in simulated environments before moving on to costly evaluations in the real world. Multi-fidelity reinforcement learning (MFRL) framework is proposed in BIB001 where multiple simulators are available. In MFRL, a cascade of simulators with increasing fidelity are used in representing state dynamics (and thus computational cost) that enables the training and validation of RL algorithms. Experiments conducted on a remote controlled car show that MFRL transfers heuristics to guide exploration in high fidelity simulators. Moreover, it allows RL algorithms to find near optimal policies for the real world with fewer expensive real world samples using a remote controlled car. Besides the MFRL setup, simulators are built to model realistic perception streams from camera, LIDAR, radar and other sensor suites. Decision making simulators require much lesser fidelity in perception while focusing vehicle dynamics and modelling the environment for path planning and trajectory opmization tasks. CARLA Challenge , is a Carla simulator based autonomous driving system challenge with pre-crash scenarios characterized in a National Highway Traffic Safety Administration report . The systems are evaluated in critical scenarios such as: Ego-vehicle loses control, ego-vehicle reacts to unseen obstacle, lane change to evade slow leading vehicle among others. The scores of agents are evaluated as a function of the aggregated distance travelled in different circuits, and total points discounted due to infractions. Recent large scale data collection on human-driven cars have lead to a data driven approach using time series data available from the GPU and IMU which were later used to extract driving primitives using unsupervised learning methods such as clustering or Bayesian optimisation BIB002 .

Deep Reinforcement Learning for Autonomous Driving: A Survey <s> D. LfD and IRL for AD applications <s> It is expected that autonomous vehicles capable of driving without human supervision will be released to market within the next decade. For user acceptance, such vehicles should not only be safe and reliable, they should also provide a comfortable user experience. However, individual perception of comfort may vary considerably among users. Whereas some users might prefer sporty driving with high accelerations, others might prefer a more relaxed style. Typically, a large number of parameters such as acceleration profiles, distances to other cars, speed during lane changes, etc., characterize a human driver's style. Manual tuning of these parameters may be a tedious and error-prone task. Therefore, we propose a learning from demonstration approach that allows the user to simply demonstrate the desired style by driving the car manually. We model the individual style in terms of a cost function and use feature-based inverse reinforcement learning to find the model parameters that fit the observed style best. Once the model has been learned, it can be used to efficiently compute trajectories for the vehicle in autonomous mode. We show that our approach is capable of learning cost functions and reproducing different driving styles using data from real drivers. <s> BIB001 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> D. LfD and IRL for AD applications <s> We trained a convolutional neural network (CNN) to map raw pixels from a single front-facing camera directly to steering commands. This end-to-end approach proved surprisingly powerful. With minimum training data from humans the system learns to drive in traffic on local roads with or without lane markings and on highways. It also operates in areas with unclear visual guidance such as in parking lots and on unpaved roads. ::: The system automatically learns internal representations of the necessary processing steps such as detecting useful road features with only the human steering angle as the training signal. We never explicitly trained it to detect, for example, the outline of roads. ::: Compared to explicit decomposition of the problem, such as lane marking detection, path planning, and control, our end-to-end system optimizes all processing steps simultaneously. We argue that this will eventually lead to better performance and smaller systems. Better performance will result because the internal components self-optimize to maximize overall system performance, instead of optimizing human-selected intermediate criteria, e.g., lane detection. Such criteria understandably are selected for ease of human interpretation which doesn't automatically guarantee maximum system performance. Smaller networks are possible because the system learns to solve the problem with the minimal number of processing steps. ::: We used an NVIDIA DevBox and Torch 7 for training and an NVIDIA DRIVE(TM) PX self-driving car computer also running Torch 7 for determining where to drive. The system operates at 30 frames per second (FPS). <s> BIB002 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> D. LfD and IRL for AD applications <s> We propose an inverse reinforcement learning (IRL) approach using Deep Q-Networks to extract the rewards in problems with large state spaces. We evaluate the performance of this approach in a simulation-based autonomous driving scenario. Our results resemble the intuitive relation between the reward function and readings of distance sensors mounted at different poses on the car. We also show that, after a few learning rounds, our simulated agent generates collision-free motions and performs human-like lane change behaviour. <s> BIB003 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> D. LfD and IRL for AD applications <s> As part of a complete software stack for autonomous driving, NVIDIA has created a neural-network-based system, known as PilotNet, which outputs steering angles given images of the road ahead. PilotNet is trained using road images paired with the steering angles generated by a human driving a data-collection car. It derives the necessary domain knowledge by observing human drivers. This eliminates the need for human engineers to anticipate what is important in an image and foresee all the necessary rules for safe driving. Road tests demonstrated that PilotNet can successfully perform lane keeping in a wide variety of driving conditions, regardless of whether lane markings are present or not. ::: The goal of the work described here is to explain what PilotNet learns and how it makes its decisions. To this end we developed a method for determining which elements in the road image most influence PilotNet's steering decision. Results show that PilotNet indeed learns to recognize relevant objects on the road. ::: In addition to learning the obvious features such as lane markings, edges of roads, and other cars, PilotNet learns more subtle features that would be hard to anticipate and program by engineers, for example, bushes lining the edge of the road and atypical vehicle classes. <s> BIB004

Early work on Behavior Cloning (BC) for driving cars in , presented agents that learn form demonstrations (LfD) that tries to mimic the behavior of an expert. BC is typically implemented as a supervised learning, and accordingly, it is hard for BC to adapt to new, unseen situations. An architecture for learning a convolutional neural network, end to end, in self-driving cars domain was proposed in BIB002 , BIB004 . The CNN is trained to map raw pixels from a single front facing camera directly to steering commands. Using a relatively small training dataset from humans/experts, the system learns to drive in traffic on local roads with or without lane markings and on highways. The network learns image representations that detect the road successfully, without being explicitly trained to do so. Authors of BIB001 proposed to learn comfortable driving trajectories optimization using expert demonstration from human drivers using Maximum Entropy Inverse RL. Authors of BIB003 used DQN as the refinement step in IRL to extract the rewards, in an effort learn human-like lane change behavior.

Deep Reinforcement Learning for Autonomous Driving: A Survey <s> VI. CHALLENGES FOR REAL-WORLD RL A. Validating RL systems <s> In recent years, significant progress has been made in solving challenging problems across various domains using deep reinforcement learning (RL). Reproducing existing work and accurately judging the improvements offered by novel methods is vital to sustaining this progress. Unfortunately, reproducing results for state-of-the-art deep RL methods is seldom straightforward. In particular, non-determinism in standard benchmark environments, combined with variance intrinsic to the methods, can make reported results tough to interpret. Without significance metrics and tighter standardization of experimental reporting, it is difficult to determine whether improvements over the prior state-of-the-art are meaningful. In this paper, we investigate challenges posed by reproducibility, proper experimental techniques, and reporting procedures. We illustrate the variability in reported metrics and results when comparing against common baselines and suggest guidelines to make future results in deep RL more reproducible. We aim to spur discussion about how to ensure continued progress in the field by minimizing wasted effort stemming from results that are non-reproducible and easily misinterpreted. <s> BIB001 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> VI. CHALLENGES FOR REAL-WORLD RL A. Validating RL systems <s> In recent years self-driving vehicles have become more commonplace on public roads, with the promise of bringing safety and efficiency to modern transportation systems. Increasing the reliability of these vehicles on the road requires an extensive suite of software tests, ideally performed on highfidelity simulators, where multiple vehicles and pedestrians interact with the self-driving vehicle. It is therefore of critical importance to ensure that self-driving software is assessed against a wide range of challenging simulated driving scenarios. The state of the art in driving scenario generation, as adopted by some of the front-runners of the self-driving car industry, still relies on human input [1]. In this paper we propose to automate the process using Bayesian optimization to generate adversarial self-driving scenarios that expose poorly-engineered or poorly-trained self-driving policies, and increase the risk of collision with simulated pedestrians and vehicles. We show that by incorporating the generated scenarios into the training set of the self-driving policy, and by fine-tuning the policy using vision-based imitation learning we obtain safer self-driving behavior. <s> BIB002

Henderson et al. BIB001 described challenges in validating reinforcement learning methods focusing on policy gradient methods for continuous control algorithms such as PPO, DDPG and TRPO as well as in reproducing benchmarks. They demonstrate with real examples that implementations often have varying code-bases and different hyper-parameter values, and that unprincipled ways to estimate the top-k rollouts could lead to incoherent interpretations on the performance of the reinforcement learning algorithms, and further more on how well they generalize. Authors concluded that evaluation could be performed either on a well defined common setup or on real-world tasks. Authors in BIB002 proposed automated generation of challenging and rare driving scenarios in highfidelity photo-realistic simulators. These adversarial scenarios are automatically discovered by parameterising the behavior of pedestrians and other vehicles on the road. Moreover, it is shown that by adding these scenarios to the training data of imitation learning, the safety is increased.

Deep Reinforcement Learning for Autonomous Driving: A Survey <s> B. Bridging the simulation-reality gap <s> We introduce a new representation learning approach for domain adaptation, in which data at training and test time come from similar but different distributions. Our approach is directly inspired by the theory on domain adaptation suggesting that, for effective domain transfer to be achieved, predictions must be made based on features that cannot discriminate between the training (source) and test (target) domains. ::: ::: The approach implements this idea in the context of neural network architectures that are trained on labeled data from the source domain and unlabeled data from the target domain (no labeled target-domain data is necessary). As the training progresses, the approach promotes the emergence of features that are (i) discriminative for the main learning task on the source domain and (ii) indiscriminate with respect to the shift between the domains. We show that this adaptation behaviour can be achieved in almost any feed-forward model by augmenting it with few standard layers and a new gradient reversal layer. The resulting augmented architecture can be trained using standard backpropagation and stochastic gradient descent, and can thus be implemented with little effort using any of the deep learning packages. ::: ::: We demonstrate the success of our approach for two distinct classification problems (document sentiment analysis and image classification), where state-of-the-art domain adaptation performance on standard benchmarks is achieved. We also validate the approach for descriptor learning task in the context of person re-identification application. <s> BIB001 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> B. Bridging the simulation-reality gap <s> Collecting well-annotated image datasets to train modern machine learning algorithms is prohibitively expensive for many tasks. One appealing alternative is rendering synthetic data where ground-truth annotations are generated automatically. Unfortunately, models trained purely on rendered images fail to generalize to real images. To address this shortcoming, prior work introduced unsupervised domain adaptation algorithms that have tried to either map representations between the two domains, or learn to extract features that are domain-invariant. In this work, we approach the problem in a new light by learning in an unsupervised manner a transformation in the pixel space from one domain to the other. Our generative adversarial network (GAN)-based method adapts source-domain images to appear as if drawn from the target domain. Our approach not only produces plausible samples, but also outperforms the state-of-the-art on a number of unsupervised domain adaptation scenarios by large margins. Finally, we demonstrate that the adaptation process generalizes to object classes unseen during training. <s> BIB002 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> B. Bridging the simulation-reality gap <s> Simulations are attractive environments for training agents as they provide an abundant source of data and alleviate certain safety concerns during the training process. But the behaviours developed by agents in simulation are often specific to the characteristics of the simulator. Due to modeling error, strategies that are successful in simulation may not transfer to their real world counterparts. In this paper, we demonstrate a simple method to bridge this"reality gap."By randomizing the dynamics of the simulator during training, we are able to develop policies that are capable of adapting to very different dynamics, including ones that differ significantly from the dynamics on which the policies were trained. This adaptivity enables the policies to generalize to the dynamics of the real world without any training on the physical system. Our approach is demonstrated on an object pushing task using a robotic arm. Despite being trained exclusively in simulation, our policies are able to maintain a similar level of performance when deployed on a real robot, reliably moving an object to a desired location from random initial configurations. We explore the impact of various design decisions and show that the resulting policies are robust to significant calibration error. <s> BIB003 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> B. Bridging the simulation-reality gap <s> Reinforcement learning is considered as a promising direction for driving policy learning. However, training autonomous driving vehicle with reinforcement learning in real environment involves non-affordable trial-and-error. It is more desirable to first train in a virtual environment and then transfer to the real environment. In this paper, we propose a novel realistic translation network to make model trained in virtual environment be workable in real world. The proposed network can convert non-realistic virtual image input into a realistic one with similar scene structure. Given realistic frames as input, driving policy trained by reinforcement learning can nicely adapt to real world driving. Experiments show that our proposed virtual to real (VR) reinforcement learning (RL) works pretty well. To our knowledge, this is the first successful case of driving policy trained by reinforcement learning that can adapt to real world driving data. <s> BIB004 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> B. Bridging the simulation-reality gap <s> Instrumenting and collecting annotated visual grasping datasets to train modern machine learning algorithms can be extremely time-consuming and expensive. An appealing alternative is to use off-the-shelf simulators to render synthetic data for which ground-truth annotations are generated automatically. Unfortunately, models trained purely on simulated data often fail to generalize to the real world. We study how randomized simulated environments and domain adaptation methods can be extended to train a grasping system to grasp novel objects from raw monocular RGB images. We extensively evaluate our approaches with a total of more than 25,000 physical test grasps, studying a range of simulation conditions and domain adaptation methods, including a novel extension of pixel-level domain adaptation that we term the GraspGAN. We show that, by using synthetic data and domain adaptation, we are able to reduce the number of real-world samples needed to achieve a given level of performance by up to 50 times, using only randomly generated simulated objects. We also show that by using only unlabeled real-world data and our GraspGAN methodology, we obtain real-world grasping performance without any real-world labels that is similar to that achieved with 939,777 labeled real-world samples. <s> BIB005 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> B. Bridging the simulation-reality gap <s> Simulation can be a powerful tool for under-standing machine learning systems and designing methods to solve real-world problems. Training and evaluating methods purely in simulation is often “doomed to succeed” at the desired task in a simulated environment, but the resulting models are incapable of operation in the real world. Here we present and evaluate a method for transferring a vision-based lane following driving policy from simulation to operation on a rural road without any real-world labels. Our approach leverages recent advances in image-to-image translation to achieve domain transfer while jointly learning a single-camera control policy from simulation control labels. We assess the driving performance of this method using both open-loop regression metrics, and closed-loop performance operating an autonomous vehicle on rural and urban roads. <s> BIB006 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> B. Bridging the simulation-reality gap <s> In this letter, we deal with the reality gap from a novel perspective, targeting transferring deep reinforcement learning (DRL) policies learned in simulated environments to the real-world domain for visual control tasks. Instead of adopting the common solutions to the problem by increasing the visual fidelity of synthetic images output from simulators during the training phase, we seek to tackle the problem by translating the real-world image streams back to the synthetic domain during the deployment phase, to make the robot feel at home . We propose this as a lightweight, flexible, and efficient solution for visual control, as first, no extra transfer steps are required during the expensive training of DRL agents in simulation; second, the trained DRL agents will not be constrained to being deployable in only one specific real-world environment; and third, the policy training and the transfer operations are decoupled, and can be conducted in parallel. Besides this, we propose a simple yet effective shift loss that is agnostic to the downstream task, to constrain the consistency between subsequent frames which is important for consistent policy outputs. We validate the shift loss for artistic style transfer for videos and domain adaptation , and validate our visual control approach in indoor and outdoor robotics experiments. <s> BIB007

Training deep networks requires collecting and annotating a lot of data which is usually costly in terms of time and effort. Using simulation environments enables the collection of large training datasets. However, the simulated data usually do not have the same data distribution compared to the real data. Accordingly, models trained on simulated environments often fail to generalise well on real environments. Domain adaptation allows a machine learning model trained on samples from a source domain to generalise on a target domain. Feature-level domain adaptation focuses on learning domaininvariant features. In BIB001 , the decisions made by deep neural networks are based on features that are both discriminative and invariant to the change of domains. Pixel level domain adaptation focuses on stylizing images from the source domain to make them similar to images of the target domain, based on image conditioned GANs. In BIB002 , the model learns a transformation in the pixel space from one domain to the other, in an unsupervised way. GAN is used to adapt simulated images to look like as if drawn from the real domain. Both feature-level and pixel-level domain adaptation are combined in BIB005 , where the results indicate that including simulated data can improve the vision-based grasping system, achieving comparable performance with 50 times fewer real-world samples. Another relatively simpler method is introduced in BIB003 , by randomizing the dynamics of the simulator during training, policies are capable of generalising to different dynamics without any training on the real system. A model trained in a virtual environment is shown to be workable in real environment BIB004 . Virtual images rendered by a simulator are first segmented to scene parsing representation and then translated to synthetic realistic images by the proposed image translation network. The proposed network segments the simulated image input, and then generates a synthetic realistic images. Accordingly, the driving policy trained by reinforcement learning can be easily adapted to real environment. Authors in BIB006 addressed the issue of performing imitation learning in simulation that transfers well to images from real world. They achieved this by unsupervised domain transfer between simulated and real world images. Additionally, the auxiliary task of predicting the steering control of the vehicle is added. By jointly training on real world images translated to simulation and the controller applied over them, a controller is learns steering in the real world domain without any ground truth from the real world. Authors remark that there were no pairwise correspondences between images in the simulated training set and the unlabelled real-world image set. Similarly, BIB007 performs domain adaptation to map real world images to simulated images. In contrast to sim-to-real methods they handle the reality gap during deployment of agents in real scenarios, by adapting the real camera streams to the synthetic modality, so as to map the unfamiliar or unseen features of real images back into the simulated style, which the agents have already learned how to deal with during training in the simulation.

Deep Reinforcement Learning for Autonomous Driving: A Survey <s> Simulator Description <s> Simulators have played a critical role in robotics research as tools for quick and efficient testing of new concepts, strategies, and algorithms. To date, most simulators have been restricted to 2D worlds, and few have matured to the point where they are both highly capable and easily adaptable. Gazebo is designed to fill this niche by creating a 3D dynamic multi-robot environment capable of recreating the complex worlds that would be encountered by the next generation of mobile robots. Its open source status, fine grained control, and high fidelity place Gazebo in a unique position to become more than just a stepping stone between the drawing board and real hardware: data visualization, simulation of remote environments, and even reverse engineering of blackbox systems are all possible applications. Gazebo is developed in cooperation with the Player and Stage projects (Gerkey, B. P., et al., July 2003), (Gerkey, B. P., et al., May 2001), (Vaughan, R. T., et al., Oct. 2003), and is available from http://playerstage.sourceforge.net/gazebo/ gazebo.html. <s> BIB001 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> Simulator Description <s> The open racing car simulator (TORCS [14]), is a modern, modular, highlyportable multi-player, multi-agent car simulator. Its high degree of modularity and portability render it ideal for artificial intelligence research. Indeed, a number of research-oriented competitions and papers have already appeared that make use of the TORCS engine. The purpose of this document is to introduce the structure of TORCS to the general artificial intelligence and machine learning community and explain how it is possible to tests agents on the platform. TORCS can be used to develop artificially intelligent (AI) agents for a variety of problems. At the car level, new simulation modules can be developed, which include intelligent control systems for various car components. At the driver level, a low-level API gives detailed (but only partial) access to the simulation state. This could be used to develop anything from mid-level control systems to complex driving agents that find optimal racing lines, react successfully in unexpected situations and make good tactical race decisions. Finally, for researchers that like a challenge and are also interested in visual processing, a 3d projection interface is available. <s> BIB002 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> Simulator Description <s> We introduce CARLA, an open-source simulator for autonomous driving research. CARLA has been developed from the ground up to support development, training, and validation of autonomous urban driving systems. In addition to open-source code and protocols, CARLA provides open digital assets (urban layouts, buildings, vehicles) that were created for this purpose and can be used freely. The simulation platform supports flexible specification of sensor suites and environmental conditions. We use CARLA to study the performance of three approaches to autonomous driving: a classic modular pipeline, an end-to-end model trained via imitation learning, and an end-to-end model trained via reinforcement learning. The approaches are evaluated in controlled scenarios of increasing difficulty, and their performance is examined via metrics provided by CARLA, illustrating the platform's utility for autonomous driving research. The supplementary video can be viewed at this https URL <s> BIB003 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> Simulator Description <s> Developing and testing algorithms for autonomous vehicles in real world is an expensive and time consuming process. Also, in order to utilize recent advances in machine intelligence and deep learning we need to collect a large amount of annotated training data in a variety of conditions and environments. We present a new simulator built on Unreal Engine that offers physically and visually realistic simulations for both of these goals. Our simulator includes a physics engine that can operate at a high frequency for real-time hardware-in-the-loop (HITL) simulations with support for popular protocols (e.g. MavLink). The simulator is designed from the ground up to be extensible to accommodate new types of vehicles, hardware platforms and software protocols. In addition, the modular design enables various components to be easily usable independently in other projects. We demonstrate the simulator by first implementing a quadrotor as an autonomous vehicle and then experimentally comparing the software components with real-world flights. <s> BIB004 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> Simulator Description <s> Flow is a new computational framework, built to support a key need triggered by the rapid growth of autonomy in ground traffic: controllers for autonomous vehicles in the presence of complex nonlinear dynamics in traffic. Leveraging recent advances in deep Reinforcement Learning (RL), Flow enables the use of RL methods such as policy gradient for traffic control and enables benchmarking the performance of classical (including hand-designed) controllers with learned policies (control laws). Flow integrates traffic microsimulator SUMO with deep reinforcement learning library rllab and enables the easy design of traffic tasks, including different networks configurations and vehicle dynamics. We use Flow to develop reliable controllers for complex problems, such as controlling mixed-autonomy traffic (involving both autonomous and human-driven vehicles) in a ring road. For this, we first show that state-of-the-art hand-designed controllers excel when in-distribution, but fail to generalize; then, we show that even simple neural network policies can solve the stabilization task across density settings and generalize to out-of-distribution settings. <s> BIB005 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> Simulator Description <s> Microscopic traffic simulation is an invaluable tool for traffic research. In recent years, both the scope of research and the capabilities of the tools have been extended considerably. This article presents the latest developments concerning intermodal traffic solutions, simulator coupling and model development and validation on the example of the open source traffic simulator SUMO. <s> BIB006

CARLA BIB003 Urban simulator, Camera & LIDAR streams, with depth & semantic segmentation, Location information TORCS BIB002 Racing Simulator, Camera stream, agent positions, testing control policies for vehicles AIRSIM BIB004 Camera stream with depth and semantic segmentation, support for drones GAZEBO (ROS) BIB001 Multi-robot physics simulator employed for path planning & vehicle control in complex 2D & 3D maps SUMO BIB006 Macro-scale modelling of traffic in cities motion planning simulators are used DeepDrive [107] Driving simulator based on unreal, providing multi-camera (eight) stream with depth Constellation NVIDIA DRIVE Constellation TM simulates camera, LIDAR and radar for autonomous driving (Proprietary) MADRaS Multi-Agent Autonomous Driving Simulator built on top of TORCS Flow BIB005 Multi-Agent Traffic Control Simulator built on top of SUMO Highway-env A gym-based environment that provides a simulator for highway based road topologies Carcraft Waymo's simulation environment (Proprietary)

Deep Reinforcement Learning for Autonomous Driving: A Survey <s> C. Sample efficiency <s> The reinforcement learning paradigm is a popular way to address problems that have only limited environmental feedback, rather than correctly labeled examples, as is common in other machine learning contexts. While significant progress has been made to improve learning in a single task, the idea of transfer learning has only recently been applied to reinforcement learning tasks. The core idea of transfer is that experience gained in learning to perform one task can help improve learning performance in a related, but different, task. In this article we present a framework that classifies transfer learning methods in terms of their capabilities and goals, and then use it to survey the existing literature, as well as to suggest future directions for transfer learning work. <s> BIB001 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> C. Sample efficiency <s> We present a framework for reinforcement learning (RL) in a scenario where multiple simulators are available with decreasing amounts of fidelity to the real-world learning scenario. Our framework is designed to limit the number of samples used in each successively higher-fidelity/cost simulator by allowing the agent to choose to run trajectories at the lowest level that will still provide it with information. The approach transfers state-action Q-values from lower-fidelity models as heuristics for the “Knows What It Knows” family of RL algorithms, which is applicable over a wide range of possible dynamics and reward representations. Theoretical proofs of the framework's sample complexity are given and empirical results are demonstrated on a remote controlled car with multiple simulators. The approach allows RL algorithms to find near-optimal policies for the real world with fewer expensive real-world samples than previous transfer approaches or learning without simulators. <s> BIB002 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> C. Sample efficiency <s> This paper presents an actor-critic deep reinforcement learning agent with experience replay that is stable, sample efficient, and performs remarkably well on challenging environments, including the discrete 57-game Atari domain and several continuous control problems. To achieve this, the paper introduces several innovations, including truncated importance sampling with bias correction, stochastic dueling network architectures, and a new trust region policy optimization method. <s> BIB003 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> C. Sample efficiency <s> In recent years deep reinforcement learning (RL) systems have attained superhuman performance in a number of challenging task domains. However, a major limitation of such applications is their demand for massive amounts of training data. A critical present objective is thus to develop deep RL methods that can adapt rapidly to new tasks. In the present work we introduce a novel approach to this challenge, which we refer to as deep meta-reinforcement learning. Previous work has shown that recurrent networks can support meta-learning in a fully supervised context. We extend this approach to the RL setting. What emerges is a system that is trained using one RL algorithm, but whose recurrent dynamics implement a second, quite separate RL procedure. This second, learned RL algorithm can differ from the original one in arbitrary ways. Importantly, because it is learned, it is configured to exploit structure in the training domain. We unpack these points in a series of seven proof-of-concept experiments, each of which examines a key aspect of deep meta-RL. We consider prospects for extending and scaling up the approach, and also point out some potentially important implications for neuroscience. <s> BIB004 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> C. Sample efficiency <s> Rather than learning new control policies for each new task, it is possible, when tasks share some structure, to compose a "meta-policy" from previously learned policies. This paper reports results from experiments using Deep Reinforcement Learning on a continuous-state, discrete-action autonomous driving simulator. We explore how Deep Neural Networks can represent meta-policies that switch among a set of previously learned policies, specifically in settings where the dynamics of a new scenario are composed of a mixture of previously learned dynamics and where the state observation is possibly corrupted by sensing noise. We also report the results of experiments varying dynamics mixes, distractor policies, magnitudes/distributions of sensing noise, and obstacles. In a fully observed experiment, the meta-policy learning algorithm achieves 2.6x the reward achieved by the next best policy composition technique with 80% less exploration. In a partially observed experiment, the meta-policy learning algorithm converges after 50 iterations while a direct application of RL fails to converge even after 200 iterations. <s> BIB005 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> C. Sample efficiency <s> We view intersection handling on autonomous vehicles as a reinforcement learning problem, and study its behavior in a transfer learning setting. We show that a network trained on one type of intersection generally is not able to generalize to other intersections. However, a network that is pre-trained on one intersection and fine-tuned on another performs better on the new task compared to training in isolation. This network also retains knowledge of the prior task, even though some forgetting occurs. Finally, we show that the benefits of fine-tuning hold when transferring simulated intersection handling knowledge to a real autonomous vehicle. <s> BIB006 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> C. Sample efficiency <s> We propose an algorithm for meta-learning that is model-agnostic, in the sense that it is compatible with any model trained with gradient descent and applicable to a variety of different learning problems, including classification, regression, and reinforcement learning. The goal of meta-learning is to train a model on a variety of learning tasks, such that it can solve new learning tasks using only a small number of training samples. In our approach, the parameters of the model are explicitly trained such that a small number of gradient steps with a small amount of training data from a new task will produce good generalization performance on that task. In effect, our method trains the model to be easy to fine-tune. We demonstrate that this approach leads to state-of-the-art performance on two few-shot image classification benchmarks, produces good results on few-shot regression, and accelerates fine-tuning for policy gradient reinforcement learning with neural network policies. <s> BIB007 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> C. Sample efficiency <s> Ability to continuously learn and adapt from limited experience in nonstationary environments is an important milestone on the path towards general intelligence. In this paper, we cast the problem of continuous adaptation into the learning-to-learn framework. We develop a simple gradient-based meta-learning algorithm suitable for adaptation in dynamically changing and adversarial scenarios. Additionally, we design a new multi-agent competitive environment, RoboSumo, and define iterated adaptation games for testing various aspects of continuous adaptation strategies. We demonstrate that meta-learning enables significantly more efficient adaptation than reactive baselines in the few-shot regime. Our experiments with a population of agents that learn and compete suggest that meta-learners are the fittest. <s> BIB008 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> C. Sample efficiency <s> A generative recurrent neural network is quickly trained in an unsupervised manner to model popular reinforcement learning environments through compressed spatio-temporal representations. The world model's extracted features are fed into compact and simple policies trained by evolution, achieving state of the art results in various environments. We also train our agent entirely inside of an environment generated by its own internal world model, and transfer this policy back into the actual environment. Interactive version of paper at this https URL <s> BIB009 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> C. Sample efficiency <s> This paper considers meta-learning problems, where there is a distribution of tasks, and we would like to obtain an agent that performs well (i.e., learns quickly) when presented with a previously unseen task sampled from this distribution. We analyze a family of algorithms for learning a parameter initialization that can be fine-tuned quickly on a new task, using only first-order derivatives for the meta-learning updates. This family includes and generalizes first-order MAML, an approximation to MAML obtained by ignoring second-order derivatives. It also includes Reptile, a new algorithm that we introduce here, which works by repeatedly sampling a task, training on it, and moving the initialization towards the trained weights on that task. We expand on the results from Finn et al. showing that first-order meta-learning algorithms perform well on some well-established benchmarks for few-shot classification, and we provide theoretical analysis aimed at understanding why these algorithms work. <s> BIB010

Animals are usually able to learn new tasks in just a few trials, benefiting from their prior knowledge about the environment. However, one of the most important challenges for reinforcement learning is sample efficiency, where the learning process requires too many samples for learning a decent policy. This issue becomes more noticeable when collection of samples is expensive or even risky. For example in the case of robot control and autonomous driving. Sample efficiency is a difficult issue due to the delayed and sparse rewards found in typical settings, in addition to the large size of the state space. Accordingly, some approaches enable the agent to learn intermediate goals via reward shaping by designing a more frequent reward function to encourage the agent to learn faster from fewer samples. Efficiency can be achieved by conducting imitation learning, where the agent is learning offline an initial policy from trajectories provided by an expert. Next, the agent can self-improve by applying RL while interacting with the environment. Transfer learning is another approach for sample efficiency, which enables the reuse of previously trained policy for a source task to initialize the learning of a target task. Policy composition presented in BIB005 propose composing previously learned basis policies to be able to reuse them for a novel task, which leads to faster learning of new policies. A survey on transfer learning in RL is presented in BIB001 . Actor Critic with Experience Replay (ACER) BIB003 , is a sample-efficient policy gradient algorithm that makes use of a replay buffer, enabling it to perform more than one gradient update using each piece of sampled experience, as well as a trust region policy optimization method. Multi-fidelity reinforcement learning (MFRL) framework BIB002 showed to transfer heuristics to guide exploration in high fidelity simulators and find near optimal policies for the real world with fewer real world samples. Moreover, model-based RL agents are known to have a competitive edge over modelfree agents, in terms of sample efficiency, where the agent can plan ahead utilizing its own model of the environment. World models proposed in BIB009 are trained quickly in an unsupervised way, via a variational autoencoder (VAE), to learn a compressed spatial and temporal representation of the environment. This approach leads to learning a compact and simple policy directly from the compressed representation. Moreover, the agent can while utilizing the internal learned environment model and transfer the policy back into the actual environment. Authors in BIB006 transferred policies learnt to handle simulated intersections to real world examples between DQN agents. Meta-learning algorithms enable agents adapt to new tasks and learn new skills rapidly from small amounts of experiences, benefiting from their prior knowledge about the world. Authors of BIB004 addressed this issue through training a recurrent neural network on a training set of interrelated tasks, where the network input includes the action selected in addition to the reward received in the previous time step. Accordingly, the agent is trained to learn to exploit the structure of the problem dynamically and solve new problems by adjusting its hidden state. A similar approach for designing RL algorithms is presented in . Rather than designing a "fast" reinforcement learning algorithm, it is represented as a recurrent neural network, and learned from data. In Model-Agnostic Meta-Learning (MAML) proposed in BIB007 , the meta-learner seeks to find an initialisation for the parameters of a neural network, that can be adapted quickly for a new task using only few examples. Continuous adaptation in dynamically changing and adversarial scenarios is presented in BIB008 via a simple gradient-based meta-learning algorithm. Additionally, Reptile BIB010 is mathematically similar to first-order MAML, making it consumes less computation and memory.

Deep Reinforcement Learning for Autonomous Driving: A Survey <s> D. Exploration issues with Imitation <s> Imitation Learning, while applied successfully on many large real-world problems, is typically addressed as a standard supervised learning problem, where it is assumed the training and testing data are i.i.d.. This is not true in imitation learning as the learned policy influences the future test inputs (states) upon which it will be tested. We show that this leads to compounding errors and a regret bound that grows quadratically in the time horizon of the task. We propose two alternative algorithms for imitation learning where training occurs over several episodes of interaction. These two approaches share in common that the learner’s policy is slowly modified from executing the expert’s policy to the learned policy. We show that this leads to stronger performance guarantees and demonstrate the improved performance on two challenging problems: training a learner to play 1) a 3D racing game (Super Tux Kart) and 2) Mario Bros.; given input images from the games and corresponding actions taken by a human expert and near-optimal planner respectively. <s> BIB001 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> D. Exploration issues with Imitation <s> Our goal is to train a policy for autonomous driving via imitation learning that is robust enough to drive a real vehicle. We find that standard behavior cloning is insufficient for handling complex driving scenarios, even when we leverage a perception system for preprocessing the input and a controller for executing the output on the car: 30 million examples are still not enough. We propose exposing the learner to synthesized data in the form of perturbations to the expert's driving, which creates interesting situations such as collisions and/or going off the road. Rather than purely imitating all data, we augment the imitation loss with additional losses that penalize undesirable events and encourage progress -- the perturbations then provide an important signal for these losses and lead to robustness of the learned model. We show that the ChauffeurNet model can handle complex situations in simulation, and present ablation experiments that emphasize the importance of each of our proposed changes and show that the model is responding to the appropriate causal factors. Finally, we demonstrate the model driving a car in the real world. <s> BIB002

In imitation learning, the agent makes use of trajectories provided by an expert. However, the distribution of states the expert encounters usually does not cover all the states the trained agent may encounter during testing. Furthermore imitation assumes that the actions are independent and identically distributed (i.i.d.). One solution consists in using the Data Aggregation (DAgger) methods BIB001 where the endto-end learned policy is executed, and extracted observationaction pairs are again labelled by the expert, and aggregated to the original expert observation-action dataset. Thus, iteratively collecting training examples from both reference and trained policies explores more valuable states and solves this lack of exploration. Following work on Search-based Structured Prediction (SEARN) BIB001 , Stochastic Mixing Iterative Learning (SMILE) trains a stochastic stationary policy over several iterations and then makes use of a geometric stochastic mixing of the policies trained. In a standard imitation learning scenario, the demonstrator is required to cover sufficient states so as to avoid unseen states during test. This constraint is costly and requires frequent human intervention. More recently, Chauffeurnet BIB002 demonstrated the limits of imitation learning where even 30 million state-action samples were insufficient to learn an optimal policy that mapped bird-eye view images (states) to control (action). The authors propose the use of simulated examples which introduced perturbations, higher diversity of scenarios such as collisions and/or going off the road. The featurenet includes an agent RNN that outputs the way point, agent box position and heading at each iteration.

Deep Reinforcement Learning for Autonomous Driving: A Survey <s> E. Intrinsic Reward functions <s> We consider learning in a Markov decision process where we are not explicitly given a reward function, but where instead we can observe an expert demonstrating the task that we want to learn to perform. This setting is useful in applications (such as the task of driving) where it may be difficult to write down an explicit reward function specifying exactly how different desiderata should be traded off. We think of the expert as trying to maximize a reward function that is expressible as a linear combination of known features, and give an algorithm for learning the task demonstrated by the expert. Our algorithm is based on using "inverse reinforcement learning" to try to recover the unknown reward function. We show that our algorithm terminates in a small number of iterations, and that even though we may never recover the expert's reward function, the policy output by the algorithm will attain performance close to that of the expert, where here performance is measured with respect to the expert's unknown reward function. <s> BIB001 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> E. Intrinsic Reward functions <s> Psychologists call behavior intrinsically motivated when it is engaged in for its own sake rather than as a step toward solving a specific problem of clear practical value. But what we learn during intrinsically motivated behavior is essential for our development as competent autonomous entities able to efficiently solve a wide range of practical problems as they arise. In this paper we present initial results from a computational study of intrinsically motivated reinforcement learning aimed at allowing artificial agents to construct and extend hierarchies of reusable skills that are needed for competent autonomy. <s> BIB002 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> E. Intrinsic Reward functions <s> In many real-world scenarios, rewards extrinsic to the agent are extremely sparse, or absent altogether. In such cases, curiosity can serve as an intrinsic reward signal to enable the agent to explore its environment and learn skills that might be useful later in its life. We formulate curiosity as the error in an agent's ability to predict the consequence of its own actions in a visual feature space learned by a self-supervised inverse dynamics model. Our formulation scales to high-dimensional continuous state spaces like images, bypasses the difficulties of directly predicting pixels, and, critically, ignores the aspects of the environment that cannot affect the agent. The proposed approach is evaluated in two environments: VizDoom and Super Mario Bros. Three broad settings are investigated: 1) sparse extrinsic reward, where curiosity allows for far fewer interactions with the environment to reach the goal; 2) exploration with no extrinsic reward, where curiosity pushes the agent to explore more efficiently; and 3) generalization to unseen scenarios (e.g. new levels of the same game) where the knowledge gained from earlier experience helps the agent explore new places much faster than starting from scratch. Demo video and code available at this https URL <s> BIB003 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> E. Intrinsic Reward functions <s> Reinforcement learning algorithms rely on carefully engineering environment rewards that are extrinsic to the agent. However, annotating each environment with hand-designed, dense rewards is not scalable, motivating the need for developing reward functions that are intrinsic to the agent. Curiosity is a type of intrinsic reward function which uses prediction error as reward signal. In this paper: (a) We perform the first large-scale study of purely curiosity-driven learning, i.e. without any extrinsic rewards, across 54 standard benchmark environments, including the Atari game suite. Our results show surprisingly good performance, and a high degree of alignment between the intrinsic curiosity objective and the hand-designed extrinsic rewards of many game environments. (b) We investigate the effect of using different feature spaces for computing prediction error and show that random features are sufficient for many popular RL game benchmarks, but learned features appear to generalize better (e.g. to novel game levels in Super Mario Bros.). (c) We demonstrate limitations of the prediction-based rewards in stochastic setups. Game-play videos and code are at https://pathak22.github.io/large-scale-curiosity/ <s> BIB004

In controlled simulated environments such as games, an explicit reward signal is given to the agent along with its sensor stream. However, in real-world robotics and autonomous driving deriving, designing a good reward functions is essential so that the desired behaviour may be learned. The most common solution has been reward shaping and consists in supplying additional well designed rewards to the agent to encourage the optimization into the direction of the optimal policy. Rewards as already pointed earlier in the paper, could be estimated by inverse RL (IRL) BIB001 , which depends on expert demonstrations. In the absence of an explicit reward shaping and expert demonstrations, agents can use intrinsic rewards or intrinsic motivation BIB002 to evaluate if their actions were good or not. Authors of BIB003 define curiosity as the error in an agent's ability to predict the consequence of its own actions in a visual feature space learned by a self-supervised inverse dynamics model. In BIB004 the agent learns a next state predictor model from its experience, and uses the error of the prediction as an intrinsic reward. This enables that agent to determine what could be a useful behavior even without extrinsic rewards.

Deep Reinforcement Learning for Autonomous Driving: A Survey <s> F. Incorporating safety in DRL <s> With the development of state-of-art deep reinforcement learning, we can efficiently tackle continuous control problems. But the deep reinforcement learning method for continuous control is based on historical data, which would make unpredicted decisions in unfamiliar scenarios. Combining deep reinforcement learning and safety based control can get good performance for self-driving and collision avoidance. In this passage, we use the Deep Deterministic Policy Gradient algorithm to implement autonomous driving without vehicles around. The vehicle can learn the driving policy in a stable and familiar environment, which is efficient and reliable. Then we use the artificial potential field to design collision avoidance algorithm with vehicles around. The path tracking method is also taken into consideration. The combination of deep reinforcement learning and safety based control performs well in most scenarios. <s> BIB001 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> F. Incorporating safety in DRL <s> Using Deep Reinforcement Learning (DRL) algorithm to deal with autonomous driving tasks usually have unsatisfied performance due to lack of robustness and means to escape local optimum. In this article, we designs a Survival-Oriented Reinforcement Learning (SORL) model that tackle these problems by setting survival rather than maximize total reward as first priority. In SORL model, we model autonomous driving task as Constrained Markov Decision Process (CMDP) and introduce Negative-Avoidance Function to learn from previous failure. The SORL model greatly speed up the training process and improve the robustness of normal Deep Reinforcement Learning algorithm. <s> BIB002

Deploying an autonomous vehicle in real environments after training directly could be dangerous. Different approaches to incorporate safety into DRL algorithms are presented here. For imitation learning based systems, Safe DAgger introduces a safety policy that learns to predict the error made by a primary policy trained initially with the supervised learning approach, without querying a reference policy. An additional safe policy takes both the partial observation of a state and a primary policy as inputs, and returns a binary label indicating whether the primary policy is likely to deviate from a reference policy without querying it. Authors of addressed safety in multi-agent Reinforcement Learning for Autonomous Driving, where a balance is maintained between unexpected behavior of other drivers or pedestrians and not to be too defensive, so that normal traffic flow is achieved. While hard constraints are maintained to guarantee the safety of driving, the problem is decomposed into a composition of a policy for desires to enable comfort driving and trajectory planning. The deep reinforcement learning algorithms for control such as DDPG and safety based control are combined in BIB001 , including artificial potential field method that is widely used for robot path planning. Using TORCS environment, the DDPG is applied first for learning a driving policy in a stable and familiar environment, then policy network and safety-based control are combined to avoid collisions. It was found that combination of DRL and safety-based control performs well in most scenarios. In order to enable DRL to escape local optima, speed up the training process and avoid danger conditions or accidents, Survival-Oriented Reinforcement Learning (SORL) model is proposed in BIB002 , where survival is favored over maximizing total reward through modeling the autonomous driving problem as a constrained MDP and introducing Negative-Avoidance Function to learn from previous failure. The SORL model was found to be not sensitive to reward function and can use different DRL algorithms like DDPG. Furthermore, a comprehensive survey on safe reinforcement learning can be found in for interested readers.

Deep Reinforcement Learning for Autonomous Driving: A Survey <s> G. Open-source frameworks <s> Recent advances in Deep Reinforcement Learning and Robotics have been driven by the presence of increasingly realistic and complex simulation environments. Many of the existing platforms, however, provide either unrealistic visuals, inaccurate physics, low task complexity, or a limited capacity for interaction among artificial agents. Furthermore, many platforms lack the ability to flexibly configure the simulation, hence turning the simulation environment into a black-box from the perspective of the learning system. Here we describe a new open source toolkit for creating and interacting with simulation environments using the Unity platform: Unity ML-Agents Toolkit. By taking advantage of Unity as a simulation platform, the toolkit enables the development of learning environments which are rich in sensory and physical complexity, provide compelling cognitive challenges, and support dynamic multi-agent interaction. We detail the platform design, communication protocol, set of example environments, and variety of training scenarios made possible via the toolkit. <s> BIB001 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> G. Open-source frameworks <s> Since the recent advent of deep reinforcement learning for game play and simulated robotic control, a multitude of new algorithms have flourished. Most are model-free algorithms which can be categorized into three families: deep Q-learning, policy gradients, and Q-value policy gradients. These have developed along separate lines of research, such that few, if any, code bases incorporate all three kinds. Yet these algorithms share a great depth of common deep reinforcement learning machinery. We are pleased to share rlpyt, which implements all three algorithm families on top of a shared, optimized infrastructure, in a single repository. It contains modular implementations of many common deep RL algorithms in Python using PyTorch, a leading deep learning library. rlpyt is designed as a high-throughput code base for small- to medium-scale research in deep RL. This white paper summarizes its features, algorithms implemented, and relation to prior work, and concludes with detailed implementation and usage notes. rlpyt is available at this https URL. <s> BIB002 </s> Deep Reinforcement Learning for Autonomous Driving: A Survey <s> G. Open-source frameworks <s> This paper introduces the Behaviour Suite for Reinforcement Learning, or bsuite for short. bsuite is a collection of carefully-designed experiments that investigate core capabilities of reinforcement learning (RL) agents with two objectives. First, to collect clear, informative and scalable problems that capture key issues in the design of general and efficient learning algorithms. Second, to study agent behaviour through their performance on these shared benchmarks. To complement this effort, we open source this http URL, which automates evaluation and analysis of any agent on bsuite. This library facilitates reproducible and accessible research on the core issues in RL, and ultimately the design of superior learning algorithms. Our code is Python, and easy to use within existing projects. We include examples with OpenAI Baselines, Dopamine as well as new reference implementations. Going forward, we hope to incorporate more excellent experiments from the research community, and commit to a periodic review of bsuite from a committee of prominent researchers. <s> BIB003

Reinforcement learning results are usually difficult to reproduce and are highly sensitive to hyper-parameter choices, which are often not reported in detail. Both researchers and practitioners need to have a reliable starting point where the well known reinforcement learning algorithms are implemented, documented and well tested. OpenAI Baselines provide a set of high-quality implementations of different reinforcement learning algorithms. The main goal for these Baselines is to make it easier for the research community to replicate, refine and create good baselines to build research on top of. Moreover, Unity Machine Learning Agents Toolkit implements core RL algorithms, games, simulations environments for training RL or IL based agents BIB001 . Reinforcement Learning Coach by Intel AI Lab enables easy experimentation with state of the art RL algorithms. Coach allows simple integration of new environments. Extending and reusing existing components is enabled through the decoupling of basic RL components. Tensorflow Agents (TF-Agents) is a versatile RL library for TensorFlow where an agent encompasses two main responsibilities: defining a policy to interact with the environment, and how to learn that policy from collected experience. TF-Agents enable developers to directly using the code from the package itself. TF-Agents is well tested and provides Jupyter notebooks example. Additionally, rlpyt BIB002 implements deep Q-learning, policy gradients and Q-value policy gradients algorithm families on top of a shared, optimized infrastructure, in a single repository in Python. rlpyt is designed as a high throughput code base for small to medium scale research in deep RL. More recently, DeepMind has released the Behaviour Suite for Reinforcement Learning (bsuite) BIB003 which is a collection of experiments that investigate capabilities of agents with two objectives: First, to collect the key issues in the design of general and efficient learning algorithms; Second, to study agent behaviour through their performance. The evaluation and analysis are automated for any agent on bsuite.

Service Discovery -- A Survey and Comparison <s> INTRODUCTION <s> For the past five years, competing industries and standards developers have been hotly pursuing automatic configuration, now coined the broader term service discovery. Jini, Universal Plug and Play (UPnP), Salutation, and Service Location Protocol are among the front-runners in this new race. However, choosing service discovery as the topic of the hour goes beyond the need for plug-and-play solutions or support for the SOHO (small office/home office) user. Service discovery's potential in mobile and pervasive computing environments motivated my choice. <s> BIB001 </s> Service Discovery -- A Survey and Comparison <s> INTRODUCTION <s> Pervasive computing environments pose unique service discovery challenges. We have developed a taxonomy of existing protocols as a basis for analyzing their approaches and identifying problems and open issues relative to service discovery in pervasive computing environments. <s> BIB002

With the increasing number of services in the internet, companies' intranets, and home networks service discovery becomes an integral part of modern networked system. This process is simple if the user and service provider know each other at run time , so what is needed is an efficient mechanism which ensures high availability of services to users and applications, and high utilization of services. In this article we survey a number of service discovery approaches. Despite the existence of a number of survey papers regarding service discovery protocols BIB001 BIB002 , we believe that a comprehensive overview of techniques and open issues for service discovery is useful. The purpose of this article is to provide a comprehensive review on service discovery approaches and to compare and analyze surveyed solutions according eight prime criteria. In this manner, we use the work both as a survey and as a guide for the design of a service discovery system. The structure of the paper is as follow: section 1 defines the objective, features and techniques of existing service discovery. Based on that, section 2 provides a comprehensive survey for leading technologies in this area. Service discovery protocols discussed in section 2 will be then compared taking various criteria in section 3. Section 4 concludes this paper with a list of future research opportunities in service discovery. In Appendix 1, we present the summary of our comparison

Service Discovery -- A Survey and Comparison <s> SLP: <s> For the past five years, competing industries and standards developers have been hotly pursuing automatic configuration, now coined the broader term service discovery. Jini, Universal Plug and Play (UPnP), Salutation, and Service Location Protocol are among the front-runners in this new race. However, choosing service discovery as the topic of the hour goes beyond the need for plug-and-play solutions or support for the SOHO (small office/home office) user. Service discovery's potential in mobile and pervasive computing environments motivated my choice. <s> BIB001 </s> Service Discovery -- A Survey and Comparison <s> SLP: <s> This article surveys research in service advertising, discovery, and selection for mobile ad hoc networks and related issues. We include a categorization of service discovery architectures for MANETs and their modes of operation, presenting their merits and drawbacks. We pay particular attention to cross-layer service discovery - a special class of efficient service discovery approaches for MANETs. We also present security issues and discuss service description options, service selection mechanisms, and service-state maintenance techniques. We conclude with a summary, an outlook, and directions for future research in this area. <s> BIB002 </s> Service Discovery -- A Survey and Comparison <s> SLP: <s> Globe is looking forward to new technologies that allow them to communicate anytime, anywhere, and using any communication device. This is possible by infrastructure-less wireless networks in which nodes can communicate directly to each other (without a central coordinator) and can relay data to each other in a selforganizing fashion.. Past researches mostly concentrated on problems on low level technologies necessary for formation of adhoc and peer-peer. But the basic role is to allow mobile users to exchange data and use each other services. Many researches has been done on lower level such as routing and resource scarcity but its important to identify higher level service discovery to implement MANET for civilian applications. So we insist on necessity of higher level service discovery that is required for MANET and various existing protocols. <s> BIB003

Service location Protocol (SLP) is an open, simple, extensible, and scalable standard for service discovery developed by the IETF (Internet Engineering Task Force). It was intended to function within IP network. SLP addresses only service discovery and leaves service invocation unspecified BIB003 . The SLP architecture consists of three main components: •User Agent (UA): software entity that sends service discovery request on a user application's behalf. •Service Agent (SA): advertises the location and characteristics of services on behalf of services. •Directory Agent (DA): a central directory collects service descriptions received from SAs in its database and process discovery queries from UAs. Figure -2-SLP agents and their transactions for service discovery and registration. As shown in figure 2 , when a new service connects to the network, the SA contacts the DA to advertise its existence (Service Registration). Registration message contains: service lifetime, URL for the service, and set of descriptive attributes for the service. Both URL schemas and attributes are defined in . Registration should be refreshed periodically by the SA to indicate its continuous existence. The same when the user needs a certain service, the UA sends request message to the DA which in turn responds with message containing URLs for all services matched against the UA needs. The client can access one of the services pointed to by the returned URL. The protocol used between the client and the service is outside the scope of the SLP specification BIB002 . To perform their respective roles UA and SA have first to discover DA location. SLP provides three methods for DA discovery: static, active, and passive. In static approach: SLP agents obtain the address of the DA using DHCP; the necessary DHCP options for SLP are defined in . With active approach: SLP agent (UA/SA) sends service request to the SLP multicast group address, a DA listening on this address will respond via unicast to the requesting agent. In passive approach: DA multicasts advertisements periodically, UAs and SAs learn the DA address from the received advertisements. It is important to note that the DA is not mandatory; it is used especially in large networks to enhance scalability. In smaller network (e.g. home network, office network) there may be no real need for DA, SLP is deployed without DA. In this case, UAs send their service requests to the SLP multicast address. The SAs announcing the service will send a unicast response to the UA. Moreover, SAs announce their presence via multicast SLP provides a powerful filter that allows UAs to select the most appropriate service from among services on the network. The UA can formulate expressive queries using operators such as AND, OR, comparators (<, =,>, <=,>=) and substring BIB001 . SLP is an open source; it does not depend on any programming language and scales well in large networks. The scalability is supported by various features such as scope concept, and multiple DAs.

Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Introduction <s> In this paper, we present Google, a prototype of a large-scale search engine which makes heavy use of the structure present in hypertext. Google is designed to crawl and index the Web efficiently and produce much more satisfying search results than existing systems. The prototype with a full text and hyperlink database of at least 24million pages is available at http://google.stanford.edu/ To engineer a search engine is a challenging task. Search engines index tens to hundreds of millions of web pages involving a comparable number of distinct terms. They answer tens of millions of queries every day. Despite the importance of large-scale search engines on the web, very little academic research has been done on them. Furthermore, due to rapid advance in technology and web proliferation, creating a web search engine today is very different from 3years ago. This paper provides an in-depth description of our large-scale web search engine - the first such detailed public description we know of to date. Apart from the problems of scaling traditional search techniques to data of this magnitude, there are new technical challenges involved with using the additional information present in hypertext to produce better search results. This paper addresses this question of how to build a practical large-scale system which can exploit the additional information present in hypertext. Also we look at the problem of how to effectively deal with uncontrolled hypertext collections, where anyone can publish anything they want. <s> BIB001 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Introduction <s> The emergence of order in natural systems is a constant source of inspiration for both physical and biological sciences. While the spatial order characterizing for example the crystals has been the basis of many advances in contemporary physics, most complex systems in nature do not offer such high degree of order. Many of these systems form complex networks whose nodes are the elements of the system and edges represent the interactions between them. ::: Traditionally complex networks have been described by the random graph theory founded in 1959 by Paul Erdohs and Alfred Renyi. One of the defining features of random graphs is that they are statistically homogeneous, and their degree distribution (characterizing the spread in the number of edges starting from a node) is a Poisson distribution. In contrast, recent empirical studies, including the work of our group, indicate that the topology of real networks is much richer than that of random graphs. In particular, the degree distribution of real networks is a power-law, indicating a heterogeneous topology in which the majority of the nodes have a small degree, but there is a significant fraction of highly connected nodes that play an important role in the connectivity of the network. ::: The scale-free topology of real networks has very important consequences on their functioning. For example, we have discovered that scale-free networks are extremely resilient to the random disruption of their nodes. On the other hand, the selective removal of the nodes with highest degree induces a rapid breakdown of the network to isolated subparts that cannot communicate with each other. ::: The non-trivial scaling of the degree distribution of real networks is also an indication of their assembly and evolution. Indeed, our modeling studies have shown us that there are general principles governing the evolution of networks. Most networks start from a small seed and grow by the addition of new nodes which attach to the nodes already in the system. This process obeys preferential attachment: the new nodes are more likely to connect to nodes with already high degree. We have proposed a simple model based on these two principles wich was able to reproduce the power-law degree distribution of real networks. Perhaps even more importantly, this model paved the way to a new paradigm of network modeling, trying to capture the evolution of networks, not just their static topology. <s> BIB002 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Introduction <s> We review the recent fast progress in statistical physics of evolving networks. Interest has focused mainly on the structural properties of random complex networks in communications, biology, social sciences and economics. A number of giant artificial networks of such a kind came into existence recently. This opens a wide field for the study of their topology, evolution, and complex processes occurring in them. Such networks possess a rich set of scaling properties. A number of them are scale-free and show striking resilience against random breakdowns. In spite of large sizes of these networks, the distances between most their vertices are short -- a feature known as the ``small-world'' effect. We discuss how growing networks self-organize into scale-free structures and the role of the mechanism of preferential linking. We consider the topological and structural properties of evolving networks, and percolation in these networks. We present a number of models demonstrating the main features of evolving networks and discuss current approaches for their simulation and analytical study. Applications of the general results to particular networks in Nature are discussed. We demonstrate the generic connections of the network growth processes with the general problems of non-equilibrium physics, econophysics, evolutionary biology, etc. <s> BIB003 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Introduction <s> From the Publisher: ::: As the World Wide Web continues to expand, it becomes increasingly difficult for users to obtain information efficiently. Because most search engines read format languages such as HTML or SGML, search results reflect formatting tags more than actual page content, which is expressed in natural language. Spinning the Semantic Web describes an exciting new type of hierarchy and standardization that will replace the current "web of links" with a "web of meaning." Using a flexible set of languages and tools, the Semantic Web will make all available information--display elements, metadata, services, images, and especially content--accessible. The result will be an immense repository of information accessible for a wide range of new applications. ::: This first handbook for the Semantic Web covers, among other topics, software agents that can negotiate and collect information, markup languages that can tag many more types of information in a document, and knowledge systems that enable machines to read Web pages and determine their reliability. The truly interdisciplinary Semantic Web combines aspects of artificial intelligence, markup languages, natural language processing, information retrieval, knowledge representation, intelligent agents, and databases. <s> BIB004 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Introduction <s> i>Recommender systems are being used by an ever-increasing number of E-commerce sites to help consumers find products to purchase. What started as a novelty has turned into a serious business tool. Recommender systems use product knowledge—either hand-coded knowledge provided by experts or “mined” knowledge learned from the behavior of consumers—to guide consumers through the often-overwhelming task of locating products they will like. In this article we present an explanation of how recommender systems are related to some traditional database analysis techniques. We examine how recommender systems help E-commerce sites increase sales and analyze the recommender systems at six market-leading sites. Based on these examples, we create a taxonomy of recommender systems, including the inputs required from the consumers, the additional knowledge required from the database, the ways the recommendations are presented to consumers, the technologies used to create the recommendations, and the level of personalization of the recommendations. We identify five commonly used E-commerce recommender application models, describe several open research problems in the field of recommender systems, and examine privacy implications of recommender systems technology. <s> BIB005 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Introduction <s> We describe the TiVo television show collaborative recommendation system which has been fielded in over one million TiVo clients for four years. Over this install base, TiVo currently has approximately 100 million ratings by users over approximately 30,000 distinct TV shows and movies. TiVo uses an item-item (show to show) form of collaborative filtering which obviates the need to keep any persistent memory of each user's viewing preferences at the TiVo server. Taking advantage of TiVo's client-server architecture has produced a novel collaborative filtering system in which the server does a minimum of work and most work is delegated to the numerous clients. Nevertheless, the server-side processing is also highly scalable and parallelizable. Although we have not performed formal empirical evaluations of its accuracy, internal studies have shown its recommendations to be useful even for multiple user households. TiVo's architecture also allows for throttling of the server so if more server-side resources become available, more correlations can be computed on the server allowing TiVo to make recommendations for niche audiences. <s> BIB006 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Introduction <s> Recommender systems are being widely applied in many application settings to suggest products, services, and information items to potential consumers. Collaborative filtering, the most successful recommendation approach, makes recommendations based on past transactions and feedback from consumers sharing similar interests. A major problem limiting the usefulness of collaborative filtering is the sparsity problem, which refers to a situation in which transactional or feedback data is sparse and insufficient to identify similarities in consumer interests. In this article, we propose to deal with this sparsity problem by applying an associative retrieval framework and related spreading activation algorithms to explore transitive associations among consumers through their past transactions and feedback. Such transitive associations are a valuable source of information to help infer consumer interests and can be explored to deal with the sparsity problem. To evaluate the effectiveness of our approach, we have conducted an experimental study using a data set from an online bookstore. We experimented with three spreading activation algorithms including a constrained Leaky Capacitor algorithm, a branch-and-bound serial symbolic search algorithm, and a Hopfield net parallel relaxation search algorithm. These algorithms were compared with several collaborative filtering approaches that do not consider the transitive associations: a simple graph search approach, two variations of the user-based approach, and an item-based approach. Our experimental results indicate that spreading activation-based approaches significantly outperformed the other collaborative filtering methods as measured by recommendation precision, recall, the F-measure, and the rank score. We also observed the over-activation effect of the spreading activation approach, that is, incorporating transitive associations with past transactional data that is not sparse may "dilute" the data used to infer user preferences and lead to degradation in recommendation performance. <s> BIB007 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Introduction <s> Recommender systems have been evaluated in many, often incomparable, ways. In this article, we review the key decisions in evaluating collaborative filtering recommender systems: the user tasks being evaluated, the types of analysis and datasets being used, the ways in which prediction quality is measured, the evaluation of prediction attributes other than quality, and the user-based evaluation of the system as a whole. In addition to reviewing the evaluation strategies used by prior researchers, we present empirical results from the analysis of various accuracy metrics on one content domain where all the tested metrics collapsed roughly into three equivalence classes. Metrics within each equivalency class were strongly correlated, while metrics from different equivalency classes were uncorrelated. <s> BIB008 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Introduction <s> This paper presents an overview of the field of recommender systems and describes the current generation of recommendation methods that are usually classified into the following three main categories: content-based, collaborative, and hybrid recommendation approaches. This paper also describes various limitations of current recommendation methods and discusses possible extensions that can improve recommendation capabilities and make recommender systems applicable to an even broader range of applications. These extensions include, among others, an improvement of understanding of users and items, incorporation of the contextual information into the recommendation process, support for multicriteria ratings, and a provision of more flexible and less intrusive types of recommendations. <s> BIB009 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Introduction <s> One of the key challenges in large information systems such as online shops and digital libraries is to discover the relevant knowledge from the enormous volume of information. Recommender systems can be viewed as a way of reducing large information spaces and to personalize information access by providing recommendations for information items based on prior usage. <s> BIB010 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Introduction <s> Coupled biological and chemical systems, neural networks, social interacting species, the Internet and the World Wide Web, are only a few examples of systems composed by a large number of highly interconnected dynamical units. The first approach to capture the global properties of such systems is to model them as graphs whose nodes represent the dynamical units, and whose links stand for the interactions between them. On the one hand, scientists have to cope with structural issues, such as characterizing the topology of a complex wiring architecture, revealing the unifying principles that are at the basis of real networks, and developing models to mimic the growth of a network and reproduce its structural properties. On the other hand, many relevant questions arise when studying complex networks’ dynamics, such as learning how a large ensemble of dynamical systems that interact through a complex wiring topology can behave collectively. We review the major concepts and results recently achieved in the study of the structure and dynamics of complex networks, and summarize the relevant applications of these ideas in many different disciplines, ranging from nonlinear science to biology, from statistical mechanics to medicine and engineering. © 2005 Elsevier B.V. All rights reserved. <s> BIB011 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Introduction <s> A tagging community's vocabulary of tags forms the basis for social navigation and shared expression.We present a user-centric model of vocabulary evolution in tagging communities based on community influence and personal tendency. We evaluate our model in an emergent tagging system by introducing tagging features into the MovieLens recommender system.We explore four tag selection algorithms for displaying tags applied by other community members. We analyze the algorithms 'effect on vocabulary evolution, tag utility, tag adoption, and user satisfaction. <s> BIB012 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Introduction <s> One-mode projecting is extensively used to compress bipartite networks. Since one-mode projection is always less informative than the bipartite representation, a proper weighting method is required to better retain the original information. In this article, inspired by the network-based resource-allocation dynamics, we raise a weighting method which can be directly applied in extracting the hidden information of networks, with remarkably better performance than the widely used global ranking method as well as collaborative filtering. This work not only provides a creditable method for compressing bipartite networks, but also highlights a possible way for the better solution of a long-standing challenge in modern information science: How to do a personal recommendation. <s> BIB013 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Introduction <s> Abstract ::: Collaborative tagging has been quickly gaining ground because of its ability to recruit the activity of web users into effectively organizing and sharing vast amounts of information. Here we collect data from a popular system and investigate the statistical properties of tag cooccurrence. We introduce a stochastic model of user behavior embodying two main aspects of collaborative tagging: (i) a frequency-bias mechanism related to the idea that users are exposed to each other's tagging activity; (ii) a notion of memory, or aging of resources, in the form of a heavy-tailed access to the past state of the system. Remarkably, our simple modeling is able to account quantitatively for the observed experimental features with a surprisingly high accuracy. This points in the direction of a universal behavior of users who, despite the complexity of their own cognitive processes and the uncoordinated and selfish nature of their tagging activity, appear to follow simple activity patterns. ::: ::: online social communities ::: statistical physics ::: social bookmarking ::: information dynamics <s> BIB014 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Introduction <s> Keywords in scientific articles have found their significance in information filtering and classification. In this article, we empirically investigated statistical characteristics and evolutionary properties of keywords in a very famous journal, namely Proceedings of the National Academy of Science of the United States of America (PNAS), including frequency distribution, temporal scaling behavior, and decay factor. The empirical results indicate that the keyword frequency in PNAS approximately follows a Zipf's law with exponent 0.86. In addition, there is a power-low correlation between the cumulative number of distinct keywords and the cumulative number of keyword occurrences. Extensive empirical analysis on some other journals' data is also presented, with decaying trends of most popular keywords being monitored. Interestingly, top journals from various subjects share very similar decaying tendency, while the journals of low impact factors exhibit completely different behavior. Those empirical characters may shed some light on the in-depth understanding of semantic evolutionary behaviors. In addition, the analysis of keyword-based system is helpful for the design of corresponding recommender systems. <s> BIB015 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Introduction <s> The enormous increase of popularity and use of the worldwide web has led in the recent years to important changes in the ways people communicate. An interesting example of this fact is provided by the now very popular social annotation systems, through which users annotate resources (such as web pages or digital photographs) with keywords known as “tags.” Understanding the rich emergent structures resulting from the uncoordinated actions of users calls for an interdisciplinary effort. In particular concepts borrowed from statistical physics, such as random walks (RWs), and complex networks theory, can effectively contribute to the mathematical modeling of social annotation systems. Here, we show that the process of social annotation can be seen as a collective but uncoordinated exploration of an underlying semantic space, pictured as a graph, through a series of RWs. This modeling framework reproduces several aspects, thus far unexplained, of social annotation, among which are the peculiar growth of the size of the vocabulary used by the community and its complex network structure that represents an externalization of semantic structures grounded in cognition and that are typically hard to access. <s> BIB016 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Introduction <s> Background: Zipf's law and Heaps' law are observed in disparate complex systems. Of particular interests, these two laws often appear together. Many theoretical models and analyses are performed to understand their co-occurrence in real systems, but it still lacks a clear picture about their relation. Methodology/Principal Findings: We show that the Heaps' law can be considered as a derivative phenomenon if the system obeys the Zipf's law. Furthermore, we refine the known approximate solution of the Heaps' exponent provided the Zipf's exponent. We show that the approximate solution is indeed an asymptotic solution for infinite systems, while in the finite-size system the Heaps' exponent is sensitive to the system size. Extensive empirical analysis on tens of disparate systems demonstrates that our refined results can better capture the relation between the Zipf's and Heaps' exponents. Conclusions/Significance: The present analysis provides a clear picture about the relation between the Zipf's law and Heaps' law without the help of any specific stochastic model, namely the Heaps' law is indeed a derivative phenomenon from Zipf's law. The presented numerical method gives considerably better estimation of the Heaps' exponent given the Zipf's exponent and the system size. Our analysis provides some insights and implications of real complex systems, for example, one can naturally obtained a better explanation of the accelerated growth of scale-free networks. <s> BIB017

The last few years have witnessed an explosion of information that the exponential growth of the Internet and World Wide Web BIB004 confronts us with an information overload: there are too much data and sources to be able to be found out those most relevant for us. Indeed, we have to make choices from thousands of movies, millions of books, billions of web pages, and so on. Evaluating all these alternatives by ourselves is not feasible at all. As a consequence, an urgent problem is how to automatically find out the relevant items for us. Internet search engine BIB001 , with the help of keyword-based queries, is an essential tool in getting what we want from the web. However, the search engine does not take into account personalization and returns the same results for people with far different habits. In addition, not all needs or tastes can be easily presented by keywords. Comparatively, recommender system , which adopts knowledge discovery techniques to provide personalized recommendations, is now considered to be the most promising way to efficiently filter out the overload information. Thus far, recommender systems have successfully found applications in e-commerce BIB005 , such as book recommendations in Amazon.com , movie recommendations in Netflix.com , video recommendations in TiVo.com BIB006 , and so on. A recommender system is able to automatically provide personalized recommendations based on the historical record of users' activities. These activities are usually represented by the connections in a user-item bipartite graph BIB007 BIB013 . So far, collaborative filtering (CF) is the most successful technique in the design of recommender systems BIB008 , where a user will be recommended items that people with similar tastes and preferences liked in the past. Despite its success, the performance of CF is strongly limited by the sparsity of data resulted from: (i) the huge number of items is far beyond user's ability to evaluate even a small fraction of them; (ii) users do not incentively wish to rate the purchased/viewed items . Besides the fundamental user-item relations, some accessorial information can be exploited to improve the algorithmic accuracy BIB009 . User profiles, usually including age, sex, nationality, job, etc., can be treated as prior known information to filter out possibly irrelevant recommendations , however, the applications are mostly forbidden or strongly restricted to respect personal privacy. Attribute-aware method BIB010 takes into account item attributes, which are defined by domain experts. Yet it is limited to the attribute vocabulary, and, on the other hand, attributes describe global properties of items which are essentially not helpful to generate personalized recommendations. In addition, content-based algorithms can provide very accurate recommendations , however, they are only effective if the items contain rich content information that can be automatically extracted out, for example, these methods are suitable for recommending books, articles and bookmarks, but not for videos, tracks or pictures. Recently, the network theory provides us a powerful and versatile tool to recognize and analyze such relation-based complex systems where nodes represent individuals, and links denote the relations among them. Therefore, many social, biological and technological and information systems can be represented as complex networks. In addition, a vast amount of efforts has been addressed in understanding the structure, evolution and dynamics of complex networks BIB002 BIB003 BIB011 . However, the advent of Web 2.0 and its affiliated applications bring a new form of user-centric paradigm which can not be fully described by pre-existing models on neither unipartite or bipartite networks. One such example is the user-driven emerging phenomenon, folksonomy , which not only allows users to upload resources (bookmarks, photos, movies, publications, etc.) but also freely assign them with user-defined words, so-called tags. Folksonomy requires no specific skills for user to participate in, broadens the semantic relations among users and resources, and eventually achieves its immediate success in a few years. Presently, a large number of such applications can be found online, such as Del.icio.us (with tags of bookmarks by users), MovieLens (with ratings of movies by users), CiteULike (with tags of publications by users), BibSonomy (with tags of bookmarks and references by users), Flickr (with tags of images by 3 users), Last.fm (with tags of music by users) etc. From the view of physics, all these online systems have performed similar statistical properties, e.g. Zipf's law like rank-frequency distribution BIB014 and Heaps' laws growth phenomenon BIB016 , between which the in-depth understanding are studied in recent works BIB015 BIB017 . With the help of those platforms, users can not only store their own resources and manage them with social tags, but also look into other users' collections to find what they might be interested in by simply keeping track of the baskets with tags. Unlike traditional information management methods where words (or indices) are normally pre-defined by experts or administrators, e.g. the library classification systems. A tagging system allows users to create arbitrary tags that even do not exist in dictionaries. Therefore, those user-defined tags can reflect user behaviors and preferences with which users can easily make acquaintance, collaborate and eventually form communities with others who have similar interests BIB012 .

Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Overview of Tag-based Recommender Systems <s> In this paper, we present Google, a prototype of a large-scale search engine which makes heavy use of the structure present in hypertext. Google is designed to crawl and index the Web efficiently and produce much more satisfying search results than existing systems. The prototype with a full text and hyperlink database of at least 24million pages is available at http://google.stanford.edu/ To engineer a search engine is a challenging task. Search engines index tens to hundreds of millions of web pages involving a comparable number of distinct terms. They answer tens of millions of queries every day. Despite the importance of large-scale search engines on the web, very little academic research has been done on them. Furthermore, due to rapid advance in technology and web proliferation, creating a web search engine today is very different from 3years ago. This paper provides an in-depth description of our large-scale web search engine - the first such detailed public description we know of to date. Apart from the problems of scaling traditional search techniques to data of this magnitude, there are new technical challenges involved with using the additional information present in hypertext to produce better search results. This paper addresses this question of how to build a practical large-scale system which can exploit the additional information present in hypertext. Also we look at the problem of how to effectively deal with uncontrolled hypertext collections, where anyone can publish anything they want. <s> BIB001 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Overview of Tag-based Recommender Systems <s> The network structure of a hyperlinked environment can be a rich source of information about the content of the environment, provided we have effective means for understanding it. We develop a set of algorithmic tools for extracting information from the link structures of such environments, and report on experiments that demonstrate their effectiveness in a variety of context on the World Wide Web. The central issue we address within our framework is the distillation of broad search topics, through the discovery of “authorative” information sources on such topics. We propose and test an algorithmic formulation of the notion of authority, based on the relationship between a set of relevant authoritative pages and the set of “hub pages” that join them together in the link structure. Our formulation has connections to the eigenvectors of certain matrices associated with the link graph; these connections in turn motivate additional heuristrics for link-based analysis. <s> BIB002 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Overview of Tag-based Recommender Systems <s> In our work we extend the traditional bipartite model of ontologies with the social dimension, leading to a tripartite model of actors, concepts and instances. We demonstrate the application of this representation by showing how community-based semantics emerges from this model through a process of graph transformation. We illustrate ontology emergence by two case studies, an analysis of a large scale folksonomy system and a novel method for the extraction of community-based ontologies from Web pages. <s> BIB003 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Overview of Tag-based Recommender Systems <s> Social bookmark tools are rapidly emerging on the Web. In such systems users are setting up lightweight conceptual structures called folksonomies. The reason for their immediate success is the fact that no specific skills are needed for participating. At the moment, however, the information retrieval support is limited. We present a formal model and a new search algorithm for folksonomies, called FolkRank, that exploits the structure of the folksonomy. The proposed algorithm is also applied to find communities within the folksonomy and is used to structure search results. All findings are demonstrated on a large scale dataset. <s> BIB004 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Overview of Tag-based Recommender Systems <s> Collaborative tagging systems, or folksonomies, have the potential of becoming technological infrastructure to support knowledge management activities in an organization or a society. There are many challenges, however. This paper presents designs that enhance collaborative tagging systems to meet some key challenges: community identification, ontology generation, user and document recommendation. Design prototypes, evaluation methodology and selected preliminary results are presented. <s> BIB005 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Overview of Tag-based Recommender Systems <s> In order to increase and enhance user-generated content contributions, it is important to understand the factors that lead people to freely share their time and knowledge with others. <s> BIB006 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Overview of Tag-based Recommender Systems <s> Collaborative tagging describes the process by which many users add metadata in the form of keywords to shared content. In this paper, we first discuss current limitations of tagging services, second, we give an overview of existing approaches. We then describe the system on component architecture that we have implemented to minimize ambiguity due to no synonym and polysemy control in folksonomy. Finally, we give experimental results. <s> BIB007 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Overview of Tag-based Recommender Systems <s> We analyze CiteULike, an online collaborative tagging system where users bookmark and annotate scientific papers. Such a system can be naturally represented as a tripartite graph whose nodes represent papers, users and tags connected by individual tag assignments. The semantics of tags is studied here, in order to uncover the hidden relationships between tags. We find that the clustering coefficient reflects the semantical patterns among tags, providing useful ideas for the designing of more efficient methods of data classification and spam detection. <s> BIB008 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Overview of Tag-based Recommender Systems <s> As a social service in Web 2.0, folksonomy provides the users the ability to save and organize their bookmarks online with "social annotations" or "tags". Social annotations are high quality descriptors of the web pages' topics as well as good indicators of web users' interests. We propose a personalized search framework to utilize folksonomy for personalized search. Specifically, three properties of folksonomy, namely the categorization, keyword, and structure property, are explored. In the framework, the rank of a web page is decided not only by the term matching between the query and the web page's content but also by the topic matching between the user's interests and the web page's topics. In the evaluation, we propose an automatic evaluation framework based on folksonomy data, which is able to help lighten the common high cost in personalized search evaluations. A series of experiments are conducted using two heterogeneous data sets, one crawled from Del.icio.us and the other from Dogear. Extensive experimental results show that our personalized search approach can significantly improve the search quality. <s> BIB009 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Overview of Tag-based Recommender Systems <s> Collaborative tagging has become an increasingly popular means for sharing and organizing Web resources, leading to a huge amount of user generated metadata. These tags represent quite a few different aspects of the resources they describe and it is not obvious whether and how these tags or subsets of them can be used for search. This paper is the first to present an in-depth study of tagging behavior for very different kinds of resources and systems - Web pages (Del.icio.us), music (Last.fm), and images (Flickr) - and compares the results with anchor text characteristics. We analyze and classify sample tags from these systems, to get an insight into what kinds of tags are used for different resources, and provide statistics on tag distributions in all three tagging environments. Since even relevant tags may not add new information to the search procedure, we also check overlap of tags with content, with metadata assigned by experts and from other sources. We discuss the potential of different kinds of tags for improving search, comparing them with user queries posted to search engines as well as through a user survey. The results are promising and provide more insight into both the use of different kinds of tags for improving search and possible extensions of tagging systems to support the creation of potentially search-relevant tags. <s> BIB010 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Overview of Tag-based Recommender Systems <s> Collaborative tagging applications allow Internet users to annotate resources with personalized tags. The complex network created by many annotations, often called a folksonomy, permits users the freedom to explore tags, resources or even other user's profiles unbound from a rigid predefined conceptual hierarchy. However, the freedom afforded users comes at a cost: an uncontrolled vocabulary can result in tag redundancy and ambiguity hindering navigation. Data mining techniques, such as clustering, provide a means to remedy these problems by identifying trends and reducing noise. Tag clusters can also be used as the basis for effective personalized recommendation assisting users in navigation. We present a personalization algorithm for recommendation in folksonomies which relies on hierarchical tag clusters. Our basic recommendation framework is independent of the clustering method, but we use a context-dependent variant of hierarchical agglomerative clustering which takes into account the user's current navigation context in cluster selection. We present extensive experimental results on two real world dataset. While the personalization algorithm is successful in both cases, our results suggest that folksonomies encompassing only one topic domain, rather than many topics, present an easier target for recommendation, perhaps because they are more focused and often less sparse. Furthermore, context dependent cluster selection, an integral step in our personalization algorithm, demonstrates more utility for recommendation in multi-topic folksonomies than in single-topic folksonomies. This observation suggests that topic selection is an important strategy for recommendation in multi-topic folksonomies. <s> BIB011 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Overview of Tag-based Recommender Systems <s> There is a growing interest into how we represent and share tagging data in collaborative tagging systems. Conventional tags, meaning freely created tags that are not associated with a structured ontology, are not naturally suited for collaborative processes, due to linguistic and grammatical variations, as well as human typing errors. Additionally, tags reflect personal views of the world by individual users, and are not normalised for synonymy, morphology or any other mapping. Our view is that the conventional approach provides very limited semantic value for collaboration. Moreover, in cases where there is some semantic value, automatically sharing semantics via computer manipulations is extremely problematic. This paper explores these problems by discussing approaches for collaborative tagging activities at a semantic level, and presenting conceptual models for collaborative tagging activities and folksonomies. We present criteria for the comparison of existing tag ontologies and discuss their strengths and weaknesses in relation to these criteria. <s> BIB012 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Overview of Tag-based Recommender Systems <s> Social tagging is the process by which many users add metadata in the form of keywords, to annotate and categorize information items (songs, pictures, web links, products etc.). Collaborative tagging systems recommend tags to users based on what tags other users have used for the same items, aiming to develop a common consensus about which tags best describe an item. However, they fail to provide appropriate tag recommendations, because: (i) users may have different interests for an information item and (ii) information items may have multiple facets. In contrast to the current tag recommendation algorithms, our approach develops a unified framework to model the three types of entities that exist in a social tagging system: users, items and tags. These data is represented by a 3-order tensor, on which latent semantic analysis and dimensionality reduction is performed using the Higher Order Singular Value Decomposition (HOSVD) technique. We perform experimental comparison of the proposed method against two state-of-the-art tag recommendations algorithms with two real data sets (Last.fm and BibSonomy). Our results show significant improvements in terms of effectiveness measured through recall/precision. <s> BIB013 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Overview of Tag-based Recommender Systems <s> In this paper we consider the problem of item recommendation in collaborative tagging communities, so called folksonomies, where users annotate interesting items with tags. Rather than following a collaborative filtering or annotation-based approach to recommendation, we extend the probabilistic latent semantic analysis (PLSA) approach and present a unified recommendation model which evolves from item user and item tag co-occurrences in parallel. The inclusion of tags reduces known collaborative filtering problems related to overfitting and allows for higher quality recommendations. Experimental results on a large snapshot of the delicious bookmarking service show the scalability of our approach and an improved recommendation quality compared to two-mode collaborative or annotation based methods. <s> BIB014 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Overview of Tag-based Recommender Systems <s> Personalized recommender systems are confronting great challenges of accuracy, diversification and novelty, especially when the data set is sparse and lacks accessorial information, such as user profiles, item attributes and explicit ratings. Collaborative tags contain rich information about personalized preferences and item contents, and are therefore potential to help in providing better recommendations. In this paper, we propose a recommendation algorithm based on an integrated diffusion on user-item-tag tripartite graphs. We use three benchmark data sets, Del.icio.us, MovieLens and BibSonomy, to evaluate our algorithm. Experimental results demonstrate that the usage of tag information can significantly improve accuracy, diversification and novelty of recommendations. <s> BIB015 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Overview of Tag-based Recommender Systems <s> In this paper, based on the user-tag-object tripartite graphs, we propose a recommendation algorithm, which considers social tags as an important role for information retrieval. Besides its low cost of computational time, the experiment results of two real-world data sets, \emph{Del.icio.us} and \emph{MovieLens}, show it can enhance the algorithmic accuracy and diversity. Especially, it can obtain more personalized recommendation results when users have diverse topics of tags. In addition, the numerical results on the dependence of algorithmic accuracy indicates that the proposed algorithm is particularly effective for small degree objects, which reminds us of the well-known \emph{cold-start} problem in recommender systems. Further empirical study shows that the proposed algorithm can significantly solve this problem in social tagging systems with heterogeneous object degree distributions. <s> BIB016 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Overview of Tag-based Recommender Systems <s> Collaborative social tagging is a popular and convenient way to organize web resources. All tags compose into a semantic structure named as folksonomies. Automatic tag suggestions can ease tagging activities of users. Various methods have been proposed for tag suggestions, which are roughly categorized into two approaches: content-based and graph-based. In this paper we present a heat diffusion method, i.e., FolkDiffusion, to rank tags for tag suggestions. Compared to existing graph-based methods, FolkDiffusion can suggest user- and resource-specific tags and prevent from topic drift. Experiments on real online social tagging datasets show the efficiency and effectiveness of FolkDiffusion compared to existing graph-based methods. <s> BIB017

Nowadays, people are confronting huge amount of information and making much effort in searching relevant or interesting items. However, as discussed in previous section, it is impossible for individuals to filter metadata from various structures and massive number of sources, especially in a user-generated information era BIB006 . The motivation of users' contribution is straightforward: they build their own data based on which they become further involved in web-based communications. Social tagging is becoming one of most popular tools in playing important rules among various social activities. Ding et al. provided good overviews of social tagging systems with emphasis on both its social impact and ontology modeling. As a consequence, social tags can be naturally considered as kind of additional yet useful resource for designing effective recommendation algorithms. Firstly, tags are freely associated by users, which can reflect their personalized preferences. Secondly, tags express the semantic relations among items, which can help evaluating the underlying item qualities. Thirdly, the co-occurrence properties of tags can be employed to build both user communities and item clusters, which be further made use of to find relevant yet interesting items for targeted individuals. Therefore, tags provide us a promising way to solve some stubborn problems in recommender systems, e.g. the cold-start problem BIB016 . Up to date, a remarkable amount of researches have discussed how to apply tags in the domain of recommender systems. Hotho et al. BIB004 proposed a modified PageRank BIB001 algorithm, namely FolkRank, to rank tags in folksonomies with the assumption that important tags are given by important users, which is akin to HITS BIB002 algorithm in internet networks. The FolkRank is then be adopted to recommend tags . In addition, due to the user-generated property, tags are considered to have high personalized information, hence can be used to design methods for both personalized searching BIB009 and recommendation. A good overview of social bookmarking and its applications in recommender systems can be found in a recent Ph.D. thesis . However, although tags are especially useful for both organizing and searching resources, there are also many studies arguing that not all tags can benefit recommendation BIB010 because of the various limitations of tags, such as polysemy, synonymy, ambiguity BIB005 BIB007 , etc. These flaws are also the side effects of the uncontrolled vocabulary, thus it remains some open issues in tagging systems: (i) singularity vs. plurality: e.g. the words cat and cats somehow have very similar meanings, however, refer to two different words in tagging systems; (ii) polysemy vs. synonymy: e.g. the word apple may refer to a kind of fruit, while it can also indicate the well-known computer company, Apple Inc., as well as its products; on the other hand, the words mac, macintosh, and apple all point to the products of Apple Inc., however, it fails again to uncover their underlying relations in tagging systems; (iii) different online tagging systems allow users to give different formats of tags, e.g. Del.icio.us only allows words to be assigned, which subsequently results in compound words with various symbols (e.g. underline, dashline, colon, etc.), leading to an unlimited formats of metadata. Such freestyle tags additionally exemplify the explosion of observed datasets, hence interfere in the analyses of the structure and user behaviors in tagging systems. Recently, researches have devoted much effort to solve those issues. Firstly, clustering-based methods BIB011 BIB008 are proposed to alleviate the word reduction problem. Secondly, semantic methods are discussed to use ontology-based to organize tags and reveal the semantic relations among them BIB003 BIB012 . Thirdly, dimension reduction and topic-based methods are put forward to discover the latent topics BIB013 BIB014 , and graphbased methods are proposed BIB015 BIB017 to solve the sparsity problem in large-scale datasets. In the following, we firstly give the evaluation metrics measured in this survey. Secondly we summarize some of the most recent and prominent tag-aware recommendation algorithms, showing and discussing how they make use of the aforementioned representations to address the some unresolved issues in recommender systems. Basically, there are three kinds of recommendations in social tagging systems: (i) predicting friends to users; (ii) recommending items to users; (iii) pushing interesting topics (tags) to users. However, as mentioned above, the most urgent problem in information era is to filter irrelevant items for individuals, therefore, in this survey, we mainly discuss the second case, and introduce some related methods discussing (i) or (iii) if necessary. Finally, we conclude with comparison of the surveyed methods and outline some future challenges of tag-aware recommendation algorithms.

Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Tag-Aware Recommendation Models <s> Social tagging is the process by which many users add metadata in the form of keywords, to annotate and categorize information items (songs, pictures, web links, products etc.). Collaborative tagging systems recommend tags to users based on what tags other users have used for the same items, aiming to develop a common consensus about which tags best describe an item. However, they fail to provide appropriate tag recommendations, because: (i) users may have different interests for an information item and (ii) information items may have multiple facets. In contrast to the current tag recommendation algorithms, our approach develops a unified framework to model the three types of entities that exist in a social tagging system: users, items and tags. These data is represented by a 3-order tensor, on which latent semantic analysis and dimensionality reduction is performed using the Higher Order Singular Value Decomposition (HOSVD) technique. We perform experimental comparison of the proposed method against two state-of-the-art tag recommendations algorithms with two real data sets (Last.fm and BibSonomy). Our results show significant improvements in terms of effectiveness measured through recall/precision. <s> BIB001 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Tag-Aware Recommendation Models <s> In the last few years we have witnessed the emergence, primarily in on-line communities, of new types of social networks that require for their representation more complex graph structures than have been employed in the past. One example is the folksonomy, a tripartite structure of users, resources, and tags -- labels collaboratively applied by the users to the resources in order to impart meaningful structure on an otherwise undifferentiated database. Here we propose a mathematical model of such tripartite structures which represents them as random hypergraphs. We show that it is possible to calculate many properties of this model exactly in the limit of large network size and we compare the results against observations of a real folksonomy, that of the on-line photography web site Flickr. We show that in some cases the model matches the properties of the observed network well, while in others there are significant differences, which we find to be attributable to the practice of multiple tagging, i.e., the application by a single user of many tags to one resource, or one tag to many resources. <s> BIB002 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Tag-Aware Recommendation Models <s> Recent years have witnessed the emergence of a new class of social networks, which require us to move beyond previously employed representations of complex graph structures. A notable example is that of the folksonomy, an online process where users collaboratively employ tags to resources to impart structure to an otherwise undifferentiated database. In a recent paper, we proposed a mathematical model that represents these structures as tripartite hypergraphs and defined basic topological quantities of interest. In this paper, we extend our model by defining additional quantities such as edge distributions, vertex similarity and correlations as well as clustering. We then empirically measure these quantities on two real life folksonomies, the popular online photo sharing site Flickr and the bookmarking site CiteULike. We find that these systems share similar qualitative features with the majority of complex networks that have been previously studied. We propose that the quantities and methodology described here can be used as a standard tool in measuring the structure of tagged networks. <s> BIB003

Formally, a social tagging network consists of three different kind of communities: users, items and tags, which subsequently form an entry set of personalized folksonomy, personomy , each follows the form F={user, item, tag 1 , tag 2 , · · · , tag t }, where t is the number of tags assigned to this item by the very user. Correspondingly, in a recommender system, a full folksonomy can be considered in two ways: (i) to be consisted of three sets, respectively of users Consequently, each binary relation can be described by a adjacent matrix, A, A ′ and A ′′ for user-item, item-tag and user-tag relations, respectively. If U i has collected I j , we set a ij = 1, otherwise a ij = 0. Analogously, we set a ′ jk = 1 if I j has been assigned by the tag T k , and a ′ jk = 0 otherwise. Furthermore, the users' preferences on tags can be represented by a adjacent matrix A ′′ , where a ′′ ik = 1 if U i has adopted T k and a ′′ ik = 0 otherwise; (ii) a ternary BIB001 or hypergraph BIB002 BIB003 based structure: only complete ternary relation is taken into account to be existed as a real link. That is to say, each relation of (u, i, t), represented as an existing component Y = 1, if it exists in a folksonomy F, and Y = 0 otherwise.

Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Evaluation Metrics <s> Recommender systems use data on past user preferences to predict possible future likes and interests. A key challenge is that while the most useful individual recommendations are to be found among diverse niche objects, the most reliably accurate results are obtained by methods that recommend objects based on user or object similarity. In this paper we introduce a new algorithm specifically to address the challenge of diversity and show how it can be used to resolve this apparent dilemma when combined in an elegant hybrid with an accuracy-focused algorithm. By tuning the hybrid appropriately we are able to obtain, without relying on any semantic or context-specific information, simultaneous gains in both accuracy and diversity of recommendations. <s> BIB001 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Evaluation Metrics <s> Recommender systems are now popular both commercially and in the research community, where many algorithms have been suggested for providing recommendations. These algorithms typically perform differently in various domains and tasks. Therefore, it is important from the research perspective, as well as from a practical view, to be able to decide on an algorithm that matches the domain and the task of interest. The standard way to make such decisions is by comparing a number of algorithms offline using some evaluation metric. Indeed, many evaluation metrics have been suggested for comparing recommendation algorithms. The decision on the proper evaluation metric is often critical, as each metric may favor a different algorithm. In this paper we review the proper construction of offline experiments for deciding on the most appropriate algorithm. We discuss three important tasks of recommender systems, and classify a set of appropriate well known evaluation metrics for each task. We demonstrate how using an improper evaluation metric can lead to the selection of an improper algorithm for the task of interest. We also discuss other important considerations when designing offline experiments. <s> BIB002

For a traditional recommender system, each data set, E, is randomly divided into two parts to test the performance of proposed algorithms: the training set, E P , is treated as 5 known information, while the testing set, E T , is used for testing. In this survey, the training set always contains 90% of entries, and the remaining 10% of entries, constitute the testing set. In addition, each division should guarantee E T E P = Ø and E T E P = E in order to make sure no redundant information is used. Furthermore, To give solid and comprehensive evaluation of the proposed algorithm, we consider metrics of both accuracy BIB002 and diversity BIB001 to characterize the performance of recommendations.

Tag-Aware Recommender Systems: A State-of-the-art Survey <s> 1. <s> A representation and interpretation of the area under a receiver operating characteristic (ROC) curve obtained by the "rating" method, or by mathematical predictions based on patient characteristics, is presented. It is shown that in such a setting the area represents the probability that a randomly chosen diseased subject is (correctly) rated or ranked with greater suspicion than a randomly chosen non-diseased subject. Moreover, this probability of a correct ranking is the same quantity that is estimated by the already well-studied nonparametric Wilcoxon statistic. These two relationships are exploited to (a) provide rapid closed-form expressions for the approximate magnitude of the sampling variability, i.e., standard error that one uses to accompany the area under a smoothed ROC curve, (b) guide in determining the size of the sample required to provide a sufficiently reliable estimate of this area, and (c) determine how large sample sizes should be to ensure that one can statistically detect difference... <s> BIB001 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> 1. <s> The discovery and analysis of community structure in networks is a topic of considerable recent interest within the physics community, but most methods proposed so far are unsuitable for very large networks because of their computational cost. Here we present a hierarchical agglomeration algorithm for detecting community structure which is faster than many competing algorithms: its running time on a network with n vertices and m edges is O(m d log n) where d is the depth of the dendrogram describing the community structure. Many real-world networks are sparse and hierarchical, with m ~ n and d ~ log n, in which case our algorithm runs in essentially linear time, O(n log^2 n). As an example of the application of this algorithm we use it to analyze a network of items for sale on the web-site of a large online retailer, items in the network being linked if they are frequently purchased by the same buyer. The network has more than 400,000 vertices and 2 million edges. We show that our algorithm can extract meaningful communities from this network, revealing large-scale patterns present in the purchasing habits of customers. <s> BIB002 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> 1. <s> One-mode projecting is extensively used to compress bipartite networks. Since one-mode projection is always less informative than the bipartite representation, a proper weighting method is required to better retain the original information. In this article, inspired by the network-based resource-allocation dynamics, we raise a weighting method which can be directly applied in extracting the hidden information of networks, with remarkably better performance than the widely used global ranking method as well as collaborative filtering. This work not only provides a creditable method for compressing bipartite networks, but also highlights a possible way for the better solution of a long-standing challenge in modern information science: How to do a personal recommendation. <s> BIB003

Ranking Score (RS) BIB003 .-In the present case, for each entry in the testing set (i.e. a user-item pair), RS is defined as the rank of the item, divided by the number of all uncollected items for the corresponding user. Apparently, the less the RS, the more accuracy the algorithm is. RS is given by averaging over all entries in the testing set. BIB001 BIB002 .-In the present case, the area under the ROC curve, abbreviated by AUC, for a particular user is the probability that a randomly selected removed item for this user (i.e., an item in the testing set and being collected by this user) is given a higher score by our algorithm than a randomly selected uncollected item (i.e, an item irrelevant to this user in neither the training set nor the testing set). The AUC for the whole system is the average over all users. If all the scores are generated from an independent and identical distribution, AUC ≈ 0.5. Therefore, the degree to which the AUC exceeds 0.5 indicates how much better the algorithm performs than pure chance. Inter Diversity (InterD) [57, BIB003 .-InterD measures the differences of different users' recommendation lists, thus can be understood as the inter-user diversity. Denote I i R the set of recommended items for user U i , then where L = |I i R | is the length of recommendation list. In average, greater or less InterD mean respectively greater or less personalization of users' recommendation lists.

Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Network-based Models <s> One-mode projecting is extensively used to compress bipartite networks. Since one-mode projection is always less informative than the bipartite representation, a proper weighting method is required to better retain the original information. In this article, inspired by the network-based resource-allocation dynamics, we raise a weighting method which can be directly applied in extracting the hidden information of networks, with remarkably better performance than the widely used global ranking method as well as collaborative filtering. This work not only provides a creditable method for compressing bipartite networks, but also highlights a possible way for the better solution of a long-standing challenge in modern information science: How to do a personal recommendation. <s> BIB001 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Network-based Models <s> While the Semantic Web has evolved to support the meaningful exchange of heterogeneous data through shared and controlled conceptualisations, Web 2.0 has demonstrated that large-scale community tagging sites can enrich the semantic web with readily accessible and valuable knowledge. In this paper, we investigate the integration of a movies folksonomy with a semantic knowledge base about user-movie rentals. The folksonomy is used to enrich the knowledge base with descriptions and categorisations of movie titles, and user interests and opinions. Using tags harvested from the Internet Movie Database, and movie rating data gathered by Netflix, we perform experiments to investigate the question that folksonomy-generated movie tag-clouds can be used to construct better user profiles that reflect a user's level of interest in different kinds of movies, and therefore, provide a basis for prediction of their rating for a previously unseen movie. <s> BIB002 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Network-based Models <s> Recommender systems use data on past user preferences to predict possible future likes and interests. A key challenge is that while the most useful individual recommendations are to be found among diverse niche objects, the most reliably accurate results are obtained by methods that recommend objects based on user or object similarity. In this paper we introduce a new algorithm specifically to address the challenge of diversity and show how it can be used to resolve this apparent dilemma when combined in an elegant hybrid with an accuracy-focused algorithm. By tuning the hybrid appropriately we are able to obtain, without relying on any semantic or context-specific information, simultaneous gains in both accuracy and diversity of recommendations. <s> BIB003 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Network-based Models <s> Using heat conduction mechanism on a social network we develop a systematic method to predict missing values as recommendations. This method can treat very large matrices that are typical of internet communities. In particular, with an innovative, exact formulation that accommodates arbitrary boundary condition, our method is easy to use in real applications. The performance is assessed by comparing with traditional recommendation methods using real data. <s> BIB004 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Network-based Models <s> Recommender Systems (RS) aim at predicting items or ratings of items that the user are interested in. Collaborative Filtering (CF) algorithms such as user- and item-based methods are the dominant techniques applied in RS algorithms. To improve recommendation quality, metadata such as content information of items has typically been used as additional knowledge. With the increasing popularity of the collaborative tagging systems, tags could be interesting and useful information to enhance RS algorithms. Unlike attributes which are "global" descriptions of items, tags are "local" descriptions of items given by the users. To the best of our knowledge, there hasn't been any prior study on tag-aware RS. In this paper, we propose a generic method that allows tags to be incorporated to standard CF algorithms, by reducing the three-dimensional correlations to three two-dimensional correlations and then applying a fusion method to re-associate these correlations. Additionally, we investigate the effect of incorporating tags information to different CF algorithms. Empirical evaluations on three CF algorithms with real-life data set demonstrate that incorporating tags to our proposed approach provides promising and significant results. <s> BIB005 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Network-based Models <s> Personalized recommender systems are confronting great challenges of accuracy, diversification and novelty, especially when the data set is sparse and lacks accessorial information, such as user profiles, item attributes and explicit ratings. Collaborative tags contain rich information about personalized preferences and item contents, and are therefore potential to help in providing better recommendations. In this paper, we propose a recommendation algorithm based on an integrated diffusion on user-item-tag tripartite graphs. We use three benchmark data sets, Del.icio.us, MovieLens and BibSonomy, to evaluate our algorithm. Experimental results demonstrate that the usage of tag information can significantly improve accuracy, diversification and novelty of recommendations. <s> BIB006 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Network-based Models <s> Collaborative tags are playing more and more important role for the organization of information systems. In this paper, we study a personalized recommendation model making use of the ternary relations among users, objects and tags. We propose a measure of user similarity based on his preference and tagging information. Two kinds of similarities between users are calculated by using a diffusion-based process, which are then integrated for recommendation. We test the proposed method in a standard collaborative filtering framework with three metrics: ranking score, Recall and Precision, and demonstrate that it performs better than the commonly used cosine similarity. <s> BIB007 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Network-based Models <s> In this paper, based on the user-tag-object tripartite graphs, we propose a recommendation algorithm, which considers social tags as an important role for information retrieval. Besides its low cost of computational time, the experiment results of two real-world data sets, \emph{Del.icio.us} and \emph{MovieLens}, show it can enhance the algorithmic accuracy and diversity. Especially, it can obtain more personalized recommendation results when users have diverse topics of tags. In addition, the numerical results on the dependence of algorithmic accuracy indicates that the proposed algorithm is particularly effective for small degree objects, which reminds us of the well-known \emph{cold-start} problem in recommender systems. Further empirical study shows that the proposed algorithm can significantly solve this problem in social tagging systems with heterogeneous object degree distributions. <s> BIB008 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Network-based Models <s> Tags are an important information source in Web 2.0. They can be used to describe users' topic preferences as well as the content of items to make personalized recommendations. However, since tags are arbitrary words given by users, they contain a lot of noise such as tag synonyms, semantic ambiguities and personal tags. Such noise brings difficulties to improve the accuracy of item recommendations. To eliminate the noise of tags, in this paper we propose to use the multiple relationships among users, items and tags to find the semantic meaning of each tag for each user individually. With the proposed approach, the relevant tags of each item and the tag preferences of each user are determined. In addition, the user and item-based collaborative filtering combined with the content filtering approach are explored. The effectiveness of the proposed approaches is demonstrated in the experiments conducted on real world datasets collected from Amazon.com and citeULike website. <s> BIB009

Recently, there are a variety of attempts utilizing tagging information for recommendation from a perspective of graph theory, Generally, a tag-based network can be viewed as a tripartite graph which consists of three integrated bipartite graphs BIB001 or a hypergraph. Therefore, network-based methods are widely used to describe the tag-based graph. Up to date, bipartite graph has been largely applied to depict massive number of online applications. For example, users rate movies, customers comment books, individuals participate in online games, etc. In a typical bipartite graph, there are two mutually connected communities, which contrastively have no link within each community, shown in Fig. 1 . Inspired by this elegant structure, two underlying network-based methods: Probability Spreading (ProbS) BIB001 BIB003 and Heat Spreading (HeatS) BIB004 BIB003 , were proposed as a starting point to apply network theory in recommender systems. ProbS is also known as random walk (RW) in computer science and mass diffusion (MD) in physics. Given a target user U i , ProbS will generate final score of each item, f j , for her/him according to following rules: Suppose that a kind of resource is initially located on items. Each item averagely distributes its resource to all neighboring users, and then each user redistributes the received resource to all his/her collected items. The final resource vector for the target user U i , f p , after the two-step mass diffusion is: where k(U l ) = m j=1 a lj is the number of collected items for user U l , and k(I s ) = n i=1 a is is the number of neighboring users for item I s . Comparatively, HeatS works based on the reverse rules of ProbS. At each step, each target will receive resources according to how active or popular it is, while ProbS distributes resources based on its own activity or popularity. Thus, BIB006 . The tripartite graph is decomposed to user-item (black links) and item-tag (red links) bipartite graphs connected by items. For the target user U 1 , the scoring process works as: (a) firstly, highlight the items, I 1 , I 3 , I 5 , collected by the target user U 1 and mark them with unit resource. That is to say: the final resource vector for the target user U i , f h , after the two-step heat spreading is: Therefore, HeatS will depress the score of popular items and is inclined to recommend the relatively cold ones, while ProbS will enhance the scoring ability of popular items. Based on the aforementioned methods, a variety of algorithms have been proposed to add tags in order to generate better recommendation performance. Zhang et al. BIB006 firstly proposed a tag-aware diffusion-based method, considering tags as additional information, which extended the resulting paradigm as reduced bipartite graphs, known as tripartite graph. In such a graph, one kind of nodes (users, items or tags) plays as a centric role to bridge the remaining two. Fig. 2 shows an example of item-centric model. In such a graph, each item of a target user will respectively distribute to its neighboring users and tags, and then all the items in database will receive their resources from their neighboring nodes. Hence, the final resource for the target user U i , f t , after two-step mass diffusion (see Fig. 2 ), will be integrated in a linear way: where is the resource of item j received from item-tag bipartite graph, k(T l ) = m j=1 a ′ jl is the number of neighboring items for tag T l , k sl is the number of neighboring tags for item I s , and λ ∈ [0, 1] is a tunable parameter to obtain the optimal performance. Table 1 shows the corresponding AUC results for three datasets: Del.icio.us, MovieLens and BibSonomy, in which the AUC values are significantly improved by considering item-tag bipartite relation. In addition, BIB006 also experi-mentally demonstrated that the incorporation of tags can enhance the Recall results for various ranges of recommendation length. Besides the accuracy, BIB006 extensively showed that tags could also promote the recommendation diversification, hence enlarge the selection vision for users. Recently, a variety of researchers have designed tag-aware algorithms by modifying the above model. Shang et al. BIB007 proposed a usercentric diffusion-based similarity, which considered users as the communication hubs to measure the coincidence among users, and found it could obtain more accurate recommendations. In addition, the tag usage frequency were measured as edge weight in user-item bipartite networks. Shang and Zhang directly regarded the frequency as weight and applied diffusion method to improve the recommendation accuracy. Wu and Zhang viewed the tag usage patten in a document vocabulary manner and applied the inverse document frequency (TF-IDF) model to calculate the weight for user-item relations. They found this weighting method could enhance the recommendation diversity. Furthermore, Zhang el al. BIB008 took such tag usage frequency into account on the user-tag and then spread the tag-based preferences to all the corresponding tags' neighboring items. The numeric results showed it could significantly enhance the algorithmic accuracy for relatively inactive or new users, and it also found that different tag usage patterns might result in different algorithmic diversity: the more diverse topic of tags users like, the more diverse results the algorithm would generate. Consequently, two fundamental roles of tags , describing and retrieving items, were firstly found applications in recommender systems. Up to date, Liang et al. BIB009 have noticed that the above methods decomposed the user-item-tag relationships into two bipartite graphs and made recommendations, which, to some extents, ignored the remaining one binary relation (e.g. user-tag for BIB006 , user-item for BIB008 ). As a result, by further eliminating the noise of tags, they used the semantic meaning of tags to represent topic preferences of users and combined it with item preferences of users to measure user-based similarity. Subsequently, the hybrid similarity was used in a standard collaborative filtering framework to obtain better Recall results in two datasets: Amazon.com and CiteULike.com. Similar measurements of user-based and item-based similarities were also widely applied by various researches BIB002 BIB005 .

Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Tensor-based Models <s> Social tagging is the process by which many users add metadata in the form of keywords, to annotate and categorize information items (songs, pictures, web links, products etc.). Collaborative tagging systems recommend tags to users based on what tags other users have used for the same items, aiming to develop a common consensus about which tags best describe an item. However, they fail to provide appropriate tag recommendations, because: (i) users may have different interests for an information item and (ii) information items may have multiple facets. In contrast to the current tag recommendation algorithms, our approach develops a unified framework to model the three types of entities that exist in a social tagging system: users, items and tags. These data is represented by a 3-order tensor, on which latent semantic analysis and dimensionality reduction is performed using the Higher Order Singular Value Decomposition (HOSVD) technique. We perform experimental comparison of the proposed method against two state-of-the-art tag recommendations algorithms with two real data sets (Last.fm and BibSonomy). Our results show significant improvements in terms of effectiveness measured through recall/precision. <s> BIB001 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Tensor-based Models <s> This survey provides an overview of higher-order tensor decompositions, their applications, and available software. A tensor is a multidimensional or $N$-way array. Decompositions of higher-order tensors (i.e., $N$-way arrays with $N \geq 3$) have applications in psycho-metrics, chemometrics, signal processing, numerical linear algebra, computer vision, numerical analysis, data mining, neuroscience, graph analysis, and elsewhere. Two particular tensor decompositions can be considered to be higher-order extensions of the matrix singular value decomposition: CANDECOMP/PARAFAC (CP) decomposes a tensor as a sum of rank-one tensors, and the Tucker decomposition is a higher-order form of principal component analysis. There are many other tensor decompositions, including INDSCAL, PARAFAC2, CANDELINC, DEDICOM, and PARATUCK2 as well as nonnegative variants of all of the above. The N-way Toolbox, Tensor Toolbox, and Multilinear Engine are examples of software packages for working with tensors. <s> BIB002 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Tensor-based Models <s> The explosive growth of e-commerce and online environments has made the issue of information search and selection increasingly serious; users are overloaded by options to consider and they may not have the time or knowledge to personally evaluate these options. Recommender systems have proven to be a valuable way for online users to cope with the information overload and have become one of the most powerful and popular tools in electronic commerce. Correspondingly, various techniques for recommendation generation have been proposed. During the last decade, many of them have also been successfully deployed in commercial environments. Recommender Systems Handbook, an edited volume, is a multi-disciplinary effort that involves world-wide experts from diverse fields, such as artificial intelligence, human computer interaction, information technology, data mining, statistics, adaptive user interfaces, decision support systems, marketing, and consumer behavior. Theoreticians and practitioners from these fields continually seek techniques for more efficient, cost-effective and accurate recommender systems. This handbook aims to impose a degree of order on this diversity, by presenting a coherent and unified repository of recommender systems major concepts, theories, methodologies, trends, challenges and applications. Extensive artificial applications, a variety of real-world applications, and detailed case studies are included. Recommender Systems Handbook illustrates how this technology can support the user in decision-making, planning and purchasing processes. It works for well known corporations such as Amazon, Google, Microsoft and AT&T. This handbook is suitable for researchers and advanced-level students in computer science as a reference. <s> BIB003

Recently, the tensor factorization (TF) BIB002 based method has attracted increasing attention to be applied in designing recommendation algorithms in social tagging systems BIB001 BIB003 . Generally, by using tensor, a ternary relation, A = {u, i, t}, can be represented as There are also other researches that define the missing values for empty triples in which the items have never been tagged, while the negative values are set for the triples in which the items are tagged in other tensors . Fig. 3 shows the illustration of the above two definitions. Table 1 . Comparison of algorithmic accuracy, measured by the AUC. Pure U-I and Pure I-T denote the pure diffusions on user-item bipartite graphs and item-tag bipartite graphs, respectively corresponding to λ=1 and λ=0. The optimal values of λ as well as the corresponding optima of AUC are presented for comparison.

Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Data set <s> We discuss a multilinear generalization of the singular value decomposition. There is a strong analogy between several properties of the matrix and the higher-order tensor decomposition; uniqueness, link with the matrix eigenvalue decomposition, first-order perturbation effects, etc., are analyzed. We investigate how tensor symmetries affect the decomposition and propose a multilinear generalization of the symmetric eigenvalue decomposition for pair-wise symmetric tensors. <s> BIB001 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Data set <s> Kernel methods provide a powerful and unified framework for pattern discovery, motivating algorithms that can act on general types of data (e.g. strings, vectors or text) and look for general types of relations (e.g. rankings, classifications, regressions, clusters). The application areas range from neural networks and pattern recognition to machine learning and data mining. This book, developed from lectures and tutorials, fulfils two major roles: firstly it provides practitioners with a large toolkit of algorithms, kernels and solutions ready to use for standard pattern discovery problems in fields such as bioinformatics, text analysis, image analysis. Secondly it provides an easy introduction for students and researchers to the growing field of kernel-based pattern analysis, demonstrating with examples how to handcraft an algorithm or a kernel for a new specific application, and covering all the necessary conceptual and mathematical tools to do so. <s> BIB002 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Data set <s> Non-linear subspaces derived using kernel methods have been found to be superior compared to linear subspaces in modeling or classification tasks of several visual phenomena. Such kernel methods include kernel PCA, kernel DA, kernel SVD and kernel QR. Since incremental computation algorithms for these methods do not exist yet, the practicality of these methods on large datasets or online video processing is minimal. We propose an approximate incremental kernel SVD algorithm for computer vision applications that require estimation of non-linear subspaces, specifically face recognition by matching image sets obtained through long-term observations or video recordings. We extend a well-known linear subspace updating algorithm to the nonlinear case by utilizing the kernel trick, and apply a reduced set construction method to produce sparse expressions for the derived subspace basis so as to maintain constant processing speed and memory usage. Experimental results demonstrate the effectiveness of the proposed method. <s> BIB003 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Data set <s> Social tagging is the process by which many users add metadata in the form of keywords, to annotate and categorize information items (songs, pictures, web links, products etc.). Collaborative tagging systems recommend tags to users based on what tags other users have used for the same items, aiming to develop a common consensus about which tags best describe an item. However, they fail to provide appropriate tag recommendations, because: (i) users may have different interests for an information item and (ii) information items may have multiple facets. In contrast to the current tag recommendation algorithms, our approach develops a unified framework to model the three types of entities that exist in a social tagging system: users, items and tags. These data is represented by a 3-order tensor, on which latent semantic analysis and dimensionality reduction is performed using the Higher Order Singular Value Decomposition (HOSVD) technique. We perform experimental comparison of the proposed method against two state-of-the-art tag recommendations algorithms with two real data sets (Last.fm and BibSonomy). Our results show significant improvements in terms of effectiveness measured through recall/precision. <s> BIB004 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Data set <s> Tagging plays an important role in many recent websites. Recommender systems can help to suggest a user the tags he might want to use for tagging a specific item. Factorization models based on the Tucker Decomposition (TD) model have been shown to provide high quality tag recommendations outperforming other approaches like PageRank, FolkRank, collaborative filtering, etc. The problem with TD models is the cubic core tensor resulting in a cubic runtime in the factorization dimension for prediction and learning. In this paper, we present the factorization model PITF (Pairwise Interaction Tensor Factorization) which is a special case of the TD model with linear runtime both for learning and prediction. PITF explicitly models the pairwise interactions between users, items and tags. The model is learned with an adaption of the Bayesian personalized ranking (BPR) criterion which originally has been introduced for item recommendation. Empirically, we show on real world datasets that this model outperforms TD largely in runtime and even can achieve better prediction quality. Besides our lab experiments, PITF has also won the ECML/PKDD Discovery Challenge 2009 for graph-based tag recommendation. <s> BIB005 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Data set <s> Item recommendation is the task of predicting a personalized ranking on a set of items (e.g. websites, movies, products). In this paper, we investigate the most common scenario with implicit feedback (e.g. clicks, purchases). There are many methods for item recommendation from implicit feedback like matrix factorization (MF) or adaptive k-nearest-neighbor (kNN). Even though these methods are designed for the item prediction task of personalized ranking, none of them is directly optimized for ranking. In this paper we present a generic optimization criterion BPR-Opt for personalized ranking that is the maximum posterior estimator derived from a Bayesian analysis of the problem. We also provide a generic learning algorithm for optimizing models with respect to BPR-Opt. The learning method is based on stochastic gradient descent with bootstrap sampling. We show how to apply our method to two state-of-the-art recommender models: matrix factorization and adaptive kNN. Our experiments indicate that for the task of personalized ranking our optimization method outperforms the standard learning techniques for MF and kNN. The results show the importance of optimizing models for the right criterion. <s> BIB006

Pure U-I Pure I-T Optimum λ opt , is set 1, and 0 otherwise . Right panel: a (u,i,t) is set negative as the triple of which the item is tagged in other existing triples rather than a (u,i,t) . The missing values are given to other empty triples . For the purpose of recommendation, Y can be represented by three low-rank feature metrics,Û ,Î,T and one core tensor,Ĉ, shown asŶ where the core tensorĈ and the feature matricesÛ ,Î andT are the parameters to be learned and ×x is the tensor x-mode dimension multiplication factor between a tensor and a matrix . In addition, the size of feature matrices are: where k U , k I , k T are the latent dimensions of the low-rank approximations for users, items and tags, respectively. Then, recommendations can be generated aŝ where the tilde denotes the feature dimensions and the hat indicates the elements of the feature matrices. Finally, the personalized recommendations list of items or tags will be displayed to the target user in a descending order. The tensor factorization is based on Singular Value Decomposition (SVD) , with which the ternary relation can be reduced to low dimensions, hence easier to be proceeded for recommendation. BIB004 used it corresponding to a TF model optimized for square-loss where all not observed values are learned as 0s. In further, developed a unified framework to model the three types of entities. Then, the three-order tensor dimension decomposition was performed by combining Higher Order Singular Value Decomposition (HOSVD) BIB001 method and the Kernel-SVD BIB002 BIB003 smoothing technique on two real-world datasets: Last.fm and BibSonomy. The results showed improvements in Recall and Precision. proposed a better learning approach for TF models, which optimized the model parameters for the AUC values. The optimization of this model is related to Bayesian personalized ranking (BPR) proposed in BIB005 . They both tried to optimize over pairs of ranking constraints, where the former focused on AUC optimization, and the lat-ter optimized for pair classification. BIB006 discussed the relationship between them in details.

Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Topic-based Models <s> Vibratory power unit for vibrating conveyers and screens comprising an asynchronous polyphase motor, at least one pair of associated unbalanced masses disposed on the shaft of said motor, with the first mass of a pair of said unbalanced masses being rigidly fastened to said shaft and with said second mass of said pair being movably arranged relative to said first mass, means for controlling and regulating the conveying rate during conveyer operation by varying the rotational speed of said motor between predetermined minimum and maximum values, said second mass being movably outwardly by centrifugal force against the pressure of spring means, said spring means being prestressed in such a manner that said second mass is, at rotational motor speeds lower than said minimum speed, held in its initial position, and at motor speeds between said lower and upper values in positions which are radially offset with respect to the axis of said motor to an extent depending on the value of said rotational motor speed. <s> BIB001 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Topic-based Models <s> A new method for automatic indexing and retrieval is described. The approach is to take advantage of implicit higher-order structure in the association of terms with documents (“semantic structure”) in order to improve the detection of relevant documents on the basis of terms found in queries. The particular technique used is singular-value decomposition, in which a large term by document matrix is decomposed into a set of ca. 100 orthogonal factors from which the original matrix can be approximated by linear combination. Documents are represented by ca. 100 item vectors of factor weights. Queries are represented as pseudo-document vectors formed from weighted combinations of terms, and documents with supra-threshold cosine values are returned. initial tests find this completely automatic method for retrieval to be promising. <s> BIB002 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Topic-based Models <s> Abstract Stochastic substitution, the Gibbs sampler, and the sampling-importance-resampling algorithm can be viewed as three alternative sampling- (or Monte Carlo-) based approaches to the calculation of numerical estimates of marginal probability distributions. The three approaches will be reviewed, compared, and contrasted in relation to various joint probability structures frequently encountered in applications. In particular, the relevance of the approaches to calculating Bayesian posterior densities for a variety of structured models will be discussed and illustrated. <s> BIB003 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Topic-based Models <s> A number of recent studies have focused on the statistical properties of networked systems such as social networks and the Worldwide Web. Researchers have concentrated particularly on a few properties that seem to be common to many networks: the small-world property, power-law degree distributions, and network transitivity. In this article, we highlight another property that is found in many networks, the property of community structure, in which network nodes are joined together in tightly knit groups, between which there are only looser connections. We propose a method for detecting such communities, built around the idea of using centrality indices to find community boundaries. We test our method on computer-generated and real-world graphs whose community structure is already known and find that the method detects this known structure with high sensitivity and reliability. We also apply the method to two networks whose community structure is not well known—a collaboration network and a food web—and find that it detects significant and informative community divisions in both cases. <s> BIB004 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Topic-based Models <s> In this paper we consider the problem of item recommendation in collaborative tagging communities, so called folksonomies, where users annotate interesting items with tags. Rather than following a collaborative filtering or annotation-based approach to recommendation, we extend the probabilistic latent semantic analysis (PLSA) approach and present a unified recommendation model which evolves from item user and item tag co-occurrences in parallel. The inclusion of tags reduces known collaborative filtering problems related to overfitting and allows for higher quality recommendations. Experimental results on a large snapshot of the delicious bookmarking service show the scalability of our approach and an improved recommendation quality compared to two-mode collaborative or annotation based methods. <s> BIB005 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Topic-based Models <s> In this paper, we propose a scalable and real-time method for tag recommendation. We model document, words and tags using tag-LDA model, which extends Latent Dirichlet Allocation model by adding the tag variable. With tag-LDA model, we can make real-time inference about the likelihood of assigning a tag to a new document, and use the likelihood to generate recommended tags. To handle large-scale data set from the web, we implement a distributed training program, which can train the tag-LDA model in parallel with multiple machines. We use a real world blog data set containing 386,012 documents to evaluate our method. The distributed training program can handle the data set efficiently. The tags recommended by our method are 32% better than search-based collaborative filtering. <s> BIB006 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Topic-based Models <s> Tagging systems have become major infrastructures on the Web. They allow users to create tags that annotate and categorize content and share them with other users, very helpful in particular for searching multimedia content. However, as tagging is not constrained by a controlled vocabulary and annotation guidelines, tags tend to be noisy and sparse. Especially new resources annotated by only a few users have often rather idiosyncratic tags that do not reflect a common perspective useful for search. In this paper we introduce an approach based on Latent Dirichlet Allocation (LDA) for recommending tags of resources in order to improve search. Resources annotated by many users and thus equipped with a fairly stable and complete tag set are used to elicit latent topics to which new resources with only a few tags are mapped. Based on this, other tags belonging to a topic can be recommended for the new resource. Our evaluation shows that the approach achieves significantly better precision and recall than the use of association rules, suggested in previous work, and also recommends more specific tags. Moreover, extending resources with these recommended tags significantly improves search for new resources. <s> BIB007 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Topic-based Models <s> Collaborative tagging systems with user generated content have become a fundamental element of websites such as Delicious, Flickr or CiteULike. By sharing common knowledge, massively linked semantic data sets are generated that provide new challenges for data mining. In this paper, we reduce the data complexity in these systems by finding meaningful topics that serve to group similar users and serve to recommend tags or resources to users. We propose a well-founded probabilistic approach that can model every aspect of a collaborative tagging system. By integrating both user information and tag information into the well-known Latent Dirichlet Allocation framework, the developed models can be used to solve a number of important information extraction and retrieval tasks. <s> BIB008 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Topic-based Models <s> Social tagging systems provide methods for users to categorise resources using their own choice of keywords (or “tags”) without being bound to a restrictive set of predefined terms. Such systems typically provide simple tag recommendations to increase the number of tags assigned to resources. In this paper we extend the latent Dirichlet allocation topic model to include user data and use the estimated probability distributions in order to provide personalised tag suggestions to users. We describe the resulting tripartite topic model in detail and show how it can be utilised to make personalised tag suggestions. Then, using data from a large-scale, real life tagging system, test our system against several baseline methods. Our experiments show a statistically significant increase in performance of our model over all key metrics, indicating that the model could be successfully used to provide further social tagging tools such as resource suggestion and collaborative filtering. <s> BIB009 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Topic-based Models <s> Exploring community is fundamental for uncovering the connections between structure and function of complex networks and for practical applications in many disciplines such as biology and sociology. In this paper, we propose a TTR-LDA-Community model which combines the Latent Dirichlet Allocation model (LDA) and the Girvan-Newman community detection algorithm with an inference mechanism. The model is then applied to data from Delicious, a popular social tagging system, over the time period of 2005-2008. Our results show that 1) users in the same community tend to be interested in similar set of topics in all time periods; and 2) topics may divide into several sub-topics and scatter into different communities over time. We evaluate the effectiveness of our model and show that the TTR-LDA-Community model is meaningful for understanding communities and outperforms TTR-LDA and LDA models in tag prediction. <s> BIB010 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Topic-based Models <s> Detecting clusters or communities in large real-world graphs such as large social or information networks is a problem of considerable interest. In practice, one typically chooses an objective function that captures the intuition of a network cluster as set of nodes with better internal connectivity than external connectivity, and then one applies approximation algorithms or heuristics to extract sets of nodes that are related to the objective function and that "look like" good communities for the application of interest. In this paper, we explore a range of network community detection methods in order to compare them and to understand their relative performance and the systematic biases in the clusters they identify. We evaluate several common objective functions that are used to formalize the notion of a network community, and we examine several different classes of approximation algorithms that aim to optimize such objective functions. In addition, rather than simply fixing an objective and asking for an approximation to the best cluster of any size, we consider a size-resolved version of the optimization problem. Considering community quality as a function of its size provides a much finer lens with which to examine community detection algorithms, since objective functions and approximation algorithms often have non-obvious size-dependent behavior. <s> BIB011

Generally, the core challenge of recommender systems is to estimate the likelihood between users and items. In the last two decades, many efforts have been devoted to build various models to measure such probabilities in information retrieval. Deerwester et al. BIB002 proposed Latent Semantic Analysis (LSA) to use a term-document matrix describing the occurrences of terms in documents. Normally, each element in the matrix is weighted by TF-IDF revealing the importance of the very term in its corresponding documents. In addition, Hofmann introduced the Probability Latent Semantic Analysis (PLSA) to improve recommendation quality for various settings by assuming a latent lower dimensional topic model as origin of observed co-occurrence distributions. Comparing with the standard LSA, PLSA is based on a mixture decomposition derived from a latent topic model which would statistically result in a more principled approach having a solid foundation. Eq. 11 gives a formula way of PLSA where word w and document d are both generated from the latent topic z, which is chosen conditionally to the document according to P (z|d), and a word is then generated from that topic according to P (w|z). However, PLSA does not allocate the topic distribution for each document, which might potentially lose information of documents with multiple subjects. Therefore, recently, a more widely used model, Latent Dirichlet Allocation (LDA) , was proposed to overcome this issue by allowing multiple latent topics with a priori Dirichlet distribution, a conjugate prior of multinomial distribution, assigned to each single document. Besides, LDA assumes that the documents are represented as random mixtures over the latent topics, each of which is given by a distribution over words. For each document d in collection D, LDA works as (see Fig. 4 ): (i) Choose θ i from Dir(α), where i runs over the document collection; (ii) For each word w ij in document d i , choose a latent topic z ij ∼ Multinomial(θ i ) and then choose a word w ij ∼ Multinomial(β z ij ). Finally, after learning the parameters by Gibbs sampling BIB003 or expectation-maximization (EM) algorithm BIB001 , the probability of the document collection can be given as Recently, those topic-based models are applied in social tagging systems for both tag and item recommendations. In BIB005 , they proposed a PLSA-based hybrid approach unifying user-item and item-tag co-occurrence to provide better item recommendations. In these two works, they measured the co-occurrence probabilities of user-item and item-tag by summing over the latent topic variables, and then maximized the likelihood of fused scenarios. Comparatively, LDA is more widely used for tag recommendation. Xi et al. BIB006 employed LDA for eliciting topics from the words in documents, as well as the co-occurrence tags, where words and tags form independent vocabulary spaces, and then recommended tags for target documents. Krestel et al. BIB007 , on the other hand, used LDA to extract hidden topics from the available tags of items and then recommended tags from these latent topics. Bundschus et al. BIB008 integrated both user information and tag information into LDA algorithm. Its generative process extracted user specific latent topics using a Topic-Tag Model adding tags and User-Topic-Tag Model adding the user layer. It assumed that users had a multinomial distribution over topics, hence, the users' interests could be modeled by each tag assignment. Finally, they used two-step latent topic realizations (user-item based and tagbased topics) to provide personalized tag recommendations. In addition, Bundschus et al. summarized different topic modeling approaches with respect to their ability to model annotations. Different from applying Bayesian rule to decompose the joint probability of itemtag and user-tag co-occupance, Harvey et al. BIB009 introduced a similar LDA-based approach for tag recommendation by decomposing the joint probability of latent topics given the tag assignments. Furthermore, Li et al. BIB010 combined LDA and GN community detection algorithm BIB004 BIB011 to observe the topic distributions of communities, as well as community evolving over time in social tagging systems. On this basis, they found that users in the same community tended to be interested in similar topics, which would shed some lights on recommendation for groups.

Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Conclusions and Outlook <s> Suppose that one of n people knows a rumor. At the first stage, he passes the rumor to someone chosen at random; at each stage, each person already informed (“knower”) communicates the rumor to a person chosen at random and independently of all other past and present choices. Denote by $S_n $ the random number of stages before everybody is informed. How large is $S_n $ typically? Frieze and Grimmet, who introduced this problem, proved that, in probability, $S_n /( \log _2 n + \log n ) \to 1$. In this paper we show that, in fact, $S_n = \log _2 n + \log n + O( 1 )$ in probability. Our proof demonstrates that the number $I( t )$ of persons informed after t stages obeys very closely, with high probability, a deterministic equation $I( {t + 1} ) = n - ( {n - I( t )} )\exp ( - I( t )/n )$, $t\geqq 0$. A case when each knower passes the rumor to several members at every stage is also discussed. <s> BIB001 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Conclusions and Outlook <s> Investigates the class of epidemic algorithms that are commonly used for the lazy transmission of updates to distributed copies of a database. These algorithms use a simple randomized communication mechanism to ensure robustness. Suppose n players communicate in parallel rounds in each of which every player calls a randomly selected communication partner. In every round, players can generate rumors (updates) that are to be distributed among all players. Whenever communication is established between two players, each one must decide which of the rumors to transmit. The major problem is that players might not know which rumors their partners have already received. For example, a standard algorithm forwarding each rumor form the calling to the called players for /spl Theta/(ln n) rounds needs to transmit the rumor /spl Theta/(n ln n) times in order to ensure that every player finally receives the rumor with high probability. We investigate whether such a large communication overhead is inherent to epidemic algorithms. On the positive side, we show that the communication overhead can be reduced significantly. We give an algorithm using only O(n ln ln n) transmissions and O(ln n) rounds. In addition, we prove the robustness of this algorithm. On the negative side, we show that any address-oblivious algorithm needs to send /spl Omega/(n ln ln n) messages for each rumor, regardless of the number of rounds. Furthermore, we give a general lower bound showing that time and communication optimality cannot be achieved simultaneously using random phone calls, i.e. every algorithm that distributes a rumor in O(ln n) rounds needs /spl omega/(n) transmissions. <s> BIB002 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Conclusions and Outlook <s> A major attack can significantly reduce the capability to deliver services in large-scale networked information systems. In this project, we have addressed the survivability of large scale heterogeneous information systems which consist of various services provided over multiple interconnected networks with different technologies. The communications network portions of such systems are referred to as multi-networks. We specifically address the issue of survivability due to physical attacks that destroy links and nodes in multi-networks. The end goal is to support critical services in the face of a major attack by making optimum use of network resources while minimizing network congestion. This is an area which is little studied, especially for large-scale heterogeneous systems. In this paper we present an overview of our contributions in this area. <s> BIB003 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Conclusions and Outlook <s> In this chapter we describe several systems that detect emerging trends in textual data. Some of the systems are semiautomatic, requiring user input to begin processing, and others are fully automatic, producing output from the input corpus without guidance. For each Emerging Trend Detection (ETD) system we describe components including linguistic and statistical features, learning algorithms, training and test set generation, visualization, and evaluation. We also provide a brief overview of several commercial products with capabilities of detecting trends in textual data, followed by an industrial viewpoint describing the importance of trend detection tools, and an overview of how such tools are used. <s> BIB004 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Conclusions and Outlook <s> We describe online collaborative communities by tripartite networks, the nodes being persons, items and tags. We introduce projection methods in order to uncover the structures of the networks, i.e. communities of users, genre families... The structuring of the network is visualised by using a tree representation. The notion of diversity in the system is also discussed. <s> BIB005 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Conclusions and Outlook <s> What determines the timing of human actions? A big question, but the science of human dynamics is here to tackle it. And its predictions are of practical value: for example, when ISPs decide what bandwidth an institution needs, they use a model of the likely timing and activity level of the individuals. Current models assume that an individual has a well defined probability of engaging in a specific action at a given moment, but evidence that the timing of human actions does not follow this pattern (of Poisson statistics) is emerging. Instead the delay between two consecutive events is best described by a heavy-tailed (power law) distribution. Albert-Laszlo Barabasi proposes an explanation for the prevalence of this behaviour. The ‘bursty’ nature of human dynamics, he finds, is a fundamental consequence of decision making. The dynamics of many social, technological and economic phenomena are driven by individual human actions, turning the quantitative understanding of human behaviour into a central question of modern science. Current models of human dynamics, used from risk assessment to communications, assume that human actions are randomly distributed in time and thus well approximated by Poisson processes1,2,3. In contrast, there is increasing evidence that the timing of many human activities, ranging from communication to entertainment and work patterns, follow non-Poisson statistics, characterized by bursts of rapidly occurring events separated by long periods of inactivity4,5,6,7,8. Here I show that the bursty nature of human behaviour is a consequence of a decision-based queuing process9,10: when individuals execute tasks based on some perceived priority, the timing of the tasks will be heavy tailed, with most tasks being rapidly executed, whereas a few experience very long waiting times. In contrast, random or priority blind execution is well approximated by uniform inter-event statistics. These finding have important implications, ranging from resource management to service allocation, in both communications and retail. <s> BIB006 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Conclusions and Outlook <s> As the number of resources on the web exceeds by far the number of documents one can track, it becomes increasingly difficult to remain up to date on ones own areas of interest. The problem becomes more severe with the increasing fraction of multimedia data, from which it is difficult to extract some conceptual description of their contents. One way to overcome this problem are social bookmark tools, which are rapidly emerging on the web. In such systems, users are setting up lightweight conceptual structures called folksonomies, and overcome thus the knowledge acquisition bottleneck. As more and more people participate in the effort, the use of a common vocabulary becomes more and more stable. We present an approach for discovering topic-specific trends within folksonomies. It is based on a differential adaptation of the PageRank algorithm to the triadic hypergraph structure of a folksonomy. The approach allows for any kind of data, as it does not rely on the internal structure of the documents. In particular, this allows to consider different data types in the same analysis step. We run experiments on a large-scale real-world snapshot of a social bookmarking system. <s> BIB007 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Conclusions and Outlook <s> Social resource sharing systems like YouTube and del.icio.us have acquired a large number of users within the last few years. They provide rich resources for data analysis, information retrieval, and knowledge discovery applications. A first step towards this end is to gain better insights into content and structure of these systems. In this paper, we will analyse the main network characteristics of two of these systems. We consider their underlying data structures - so-called folksonomies - as tri-partite hypergraphs, and adapt classical network measures like characteristic path length and clustering coefficient to them. ::: ::: Subsequently, we introduce a network of tag co-occurrence and investigate some of its statistical properties, focusing on correlations in node connectivity and pointing out features that reflect emergent semantics within the folksonomy. We show that simple statistical indicators unambiguously spot non-social behavior such as spam. <s> BIB008 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Conclusions and Outlook <s> We consider the problem of visualizing the evolution of tags within the Flickr (flickr.com) online image sharing community. Any user of the Flickr service may append a tag to any photo in the system. Over the past year, users have on average added over a million tags each week. Understanding the evolution of these tags over time is therefore a challenging task. We present a new approach based on a characterization of the most interesting tags associated with a sliding interval of time. An animation provided via Flash in a web browser allows the user to observe and interact with the interesting tags as they evolve over time.New algorithms and data structures are required to support the efficient generation of this visualization. We combine a novel solution to an interval covering problem with extensions to previous work on score aggregation in order to create an efficient backend system capable of producing visualizations at arbitrary scales on this large dataset in real time. <s> BIB009 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Conclusions and Outlook <s> Collaborative tagging applications allow Internet users to annotate resources with personalized tags. The complex network created by many annotations, often called a folksonomy, permits users the freedom to explore tags, resources or even other user's profiles unbound from a rigid predefined conceptual hierarchy. However, the freedom afforded users comes at a cost: an uncontrolled vocabulary can result in tag redundancy and ambiguity hindering navigation. Data mining techniques, such as clustering, provide a means to remedy these problems by identifying trends and reducing noise. Tag clusters can also be used as the basis for effective personalized recommendation assisting users in navigation. We present a personalization algorithm for recommendation in folksonomies which relies on hierarchical tag clusters. Our basic recommendation framework is independent of the clustering method, but we use a context-dependent variant of hierarchical agglomerative clustering which takes into account the user's current navigation context in cluster selection. We present extensive experimental results on two real world dataset. While the personalization algorithm is successful in both cases, our results suggest that folksonomies encompassing only one topic domain, rather than many topics, present an easier target for recommendation, perhaps because they are more focused and often less sparse. Furthermore, context dependent cluster selection, an integral step in our personalization algorithm, demonstrates more utility for recommendation in multi-topic folksonomies than in single-topic folksonomies. This observation suggests that topic selection is an important strategy for recommendation in multi-topic folksonomies. <s> BIB010 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Conclusions and Outlook <s> Tagging systems allow users to interactively annotate a pool of shared resources using descriptive strings called tags. Tags are used to guide users to interesting resources and help them build communities that share their expertise and resources. As tagging systems are gaining in popularity, they become more susceptible to tag spam: misleading tags that are generated in order to increase the visibility of some resources or simply to confuse users. Our goal is to understand this problem better. In particular, we are interested in answers to questions such as: How many malicious users can a tagging system tolerate before results significantly degradeq What types of tagging systems are more vulnerable to malicious attacksq What would be the effort and the impact of employing a trusted moderator to find bad postingsq Can a system automatically protect itself from spam, for instance, by exploiting user tag patternsq In a quest for answers to these questions, we introduce a framework for modeling tagging systems and user tagging behavior. We also describe a method for ranking documents matching a tag based on taggers' reliability. Using our framework, we study the behavior of existing approaches under malicious attacks and the impact of a moderator and our ranking method. <s> BIB011 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Conclusions and Outlook <s> In this paper, we look at the "social tag prediction" problem. Given a set of objects, and a set of tags applied to those objects by users, can we predict whether a given tag could/should be applied to a particular object? We investigated this question using one of the largest crawls of the social bookmarking system del.icio.us gathered to date. For URLs in del.icio.us, we predicted tags based on page text, anchor text, surrounding hosts, and other tags applied to the URL. We found an entropy-based metric which captures the generality of a particular tag and informs an analysis of how well that tag can be predicted. We also found that tag-based association rules can produce very high-precision predictions as well as giving deeper understanding into the relationships between tags. Our results have implications for both the study of tagging systems as potential information retrieval tools, and for the design of such systems. <s> BIB012 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Conclusions and Outlook <s> Recent years have witnessed the emergence of a new class of social networks, which require us to move beyond previously employed representations of complex graph structures. A notable example is that of the folksonomy, an online process where users collaboratively employ tags to resources to impart structure to an otherwise undifferentiated database. In a recent paper, we proposed a mathematical model that represents these structures as tripartite hypergraphs and defined basic topological quantities of interest. In this paper, we extend our model by defining additional quantities such as edge distributions, vertex similarity and correlations as well as clustering. We then empirically measure these quantities on two real life folksonomies, the popular online photo sharing site Flickr and the bookmarking site CiteULike. We find that these systems share similar qualitative features with the majority of complex networks that have been previously studied. We propose that the quantities and methodology described here can be used as a standard tool in measuring the structure of tagged networks. <s> BIB013 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Conclusions and Outlook <s> Automatically clustering web pages into semantic groups promises improved search and browsing on the web. In this paper, we demonstrate how user-generated tags from large-scale social bookmarking websites such as del.icio.us can be used as a complementary data source to page text and anchor text for improving automatic clustering of web pages. This paper explores the use of tags in 1) K-means clustering in an extended vector space model that includes tags as well as page text and 2) a novel generative clustering algorithm based on latent Dirichlet allocation that jointly models text and tags. We evaluate the models by comparing their output to an established web directory. We find that the naive inclusion of tagging data improves cluster quality versus page text alone, but a more principled inclusion can substantially improve the quality of all models with a statistically significant absolute F-score increase of 4%. The generative model outperforms K-means with another 8% F-score increase. <s> BIB014 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Conclusions and Outlook <s> Most news recommender systems try to identify users' interests and news' attributes and use them to obtain recommendations. Here we propose an adaptive model which combines similarities in users' rating patterns with epidemic-like spreading of news on an evolving network. We study the model by computer agent-based simulations, measure its performance and discuss its robustness against bias and malicious behavior. Subject to the approval fraction of news recommended, the proposed model outperforms the widely adopted recommendation of news according to their absolute or relative popularity. This model provides a general social mechanism for recommender systems and may find its applications also in other types of recommendation. <s> BIB015 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Conclusions and Outlook <s> Complex hypernetworks are ubiquitous in real-life systems. While a substantial body of previous research has only focused on the applications of hypernetworks, relatively little work has investigated the evolving models of hypernetworks. Considering the formations of many real world networks, we propose two evolving mechanisms of the hyperedge growth and the hyperedge preferential attachment, then construct an evolving hypernetwork model. We introduce some basic topological quantities, such as a variety of degree distributions, clustering coefficients as well as average path length. We numerically investigate these quantities in the limit of large hypernetwork size and find that our hypernetwork model shares similar qualitative features with the majority of complex networks that have been previously studied, such as the scale-free property of the degree distribution and a high degree of clustering, as well as the small-world property. It is expected that our attempt in the hypernetwork model can bring the upsurge in the study of the hypernetwork model in further. <s> BIB016 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Conclusions and Outlook <s> Tapping into the wisdom of the crowd, social tagging can be considered an alternative mechanism - as opposed to Web search - for organizing and discovering information on the Web. Effective tag-based recommendation of information items, such as Web resources, is a critical aspect of this social information discovery mechanism. A precise understanding of the information structure of social tagging systems lies at the core of an effective tag-based recommendation method. While most of the existing research either implicitly or explicitly assumes a simple tripartite graph structure for this purpose, we propose a comprehensive information structure to capture all types of co-occurrence information in the tagging data. Based on the proposed information structure, we further propose a unified user profiling scheme to make full use of all available information. Finally, supported by our proposed user profile, we propose a novel framework for collaborative filtering in social tagging systems. In our proposed framework, we first generate joint item-tag recommendations, with tags indicating topical interests of users in target items. These joint recommendations are then refined by the wisdom from the crowd and projected to the item space for final item recommendations. Evaluation using three real-world datasets shows that our proposed recommendation approach significantly outperformed state-of-the-art approaches. <s> BIB017 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Conclusions and Outlook <s> Social recommendation, that an individual recommends an item to another, has gained popularity and success in web applications such as online sharing and shopping services. It is largely different from a traditional recommendation where an automatic system recommends an item to a user. In a social recommendation, the interpersonal influence plays a critical role but is usually ignored in traditional recommendation systems, which recommend items based on user-item utility. In this paper, we propose an approach to model the utility of a social recommendation through combining three factors, i.e. receiver interests, item qualities and interpersonal influences. In our approach, values of all factors can be learned from user behaviors. Experiments are conducted to compare our approach with three conventional methods in social recommendation prediction. Empirical results show the effectiveness of our approach, where an increase by 26% in prediction accuracy can be observed. <s> BIB018 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Conclusions and Outlook <s> Recommender systems help people cope with the problem of information overload. A recently proposed adaptive news recommender model [M. Medo, Y.-C. Zhang, T. Zhou, Europhys. Lett. 88, 38005 (2009)] is based on epidemic-like spreading of news in a social network. By means of agent-based simulations we study a “good get richer” feature of the model and determine which attributes are necessary for a user to play a leading role in the network. We further investigate the filtering efficiency of the model as well as its robustness against malicious and spamming behaviour. We show that incorporating user reputation in the recommendation process can substantially improve the outcome. <s> BIB019 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Conclusions and Outlook <s> Link prediction in complex networks has attracted increasing attention from both physical and computer science communities. The algorithms can be used to extract missing information, identify spurious interactions, evaluate network evolving mechanisms, and so on. This article summaries recent progress about link prediction algorithms, emphasizing on the contributions from physical perspectives and approaches, such as the random-walk-based methods and the maximum likelihood methods. We also introduce three typical applications: reconstruction of networks, evaluation of network evolving mechanism and classification of partially labelled networks. Finally, we introduce some applications and outline future challenges of link prediction algorithms. <s> BIB020 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Conclusions and Outlook <s> In the scope of the Challenge on Context-aware Movie Recommendation (CAMRa2010), context can mean temporal context (Task 1), mood (Task 2), or social context (Task 3). We suggest to use Pairwise Interaction Tensor Factorization (PITF), a method used for personalized tag recommendation, to model the temporal (week) context in Task 1 of the challenge. We also present an extended version of PITF that handles the week context in a smoother way. In the experiments, we compare PITF against different item recommendation baselines that do not take context into account, and a non-personalized context-aware baseline. <s> BIB021 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Conclusions and Outlook <s> Accurately capturing user preferences over time is a great practical challenge in recommender systems. Simple correlation over time is typically not meaningful, since users change their preferences due to different external events. User behavior can often be determined by individual's long-term and short-term preferences. How to represent users' long-term and short-term preferences? How to leverage them for temporal recommendation? To address these challenges, we propose Session-based Temporal Graph (STG) which simultaneously models users' long-term and short-term preferences over time. Based on the STG model framework, we propose a novel recommendation algorithm Injected Preference Fusion (IPF) and extend the personalized Random Walk for temporal recommendation. Finally, we evaluate the effectiveness of our method using two real datasets on citations and social bookmarking, in which our proposed method IPF gives 15%-34% improvement over the previous state-of-the-art. <s> BIB022 </s> Tag-Aware Recommender Systems: A State-of-the-art Survey <s> Conclusions and Outlook <s> Individuals often imitate each other to fall into the typical group, leading to a self-organized state of typical behaviors in a community. In this paper, we model self-organization in social tagging systems and illustrate the underlying interaction and dynamics. Specifically, we introduce a model in which individuals adjust their own tagging tendency to imitate the average tagging tendency. We found that when users are of low confidence, they tend to imitate others and lead to a self-organized state with active tagging. On the other hand, when users are of high confidence and are stubborn to change, tagging becomes inactive. We observe a phase transition at a critical level of user confidence when the system changes from one regime to the other. The distributions of post length obtained from the model are compared to real data, which show good agreement. <s> BIB023

In this survey, we summarized the progress of studies on tag-aware recommender systems (RS), emphasizing on the recent contributions by both statistical physicists and computer scientists in three aspects: (i) network-based methods; (ii) tensor-based methods; (iii) topicbased methods. Generally, there is no single method that can fully address all the problems existing in RS. Network-based and tensorbased methods can overcome the sparsity of large-scale data, hence can be used for designing efficient algorithms. However, they only focus on the network structure, while lack considerations of relations among tags. Comparatively, topic-based methods can distinguish tags into different topics, hence can produce more meaningful and understandable recommendations. However, since most of topicbased methods use machine learning to iteratively refine the results, they require highefficient hardwares for computation, and thus consume more computation time. Similar problem lies in tensor-based methods for dimension reduction process. Therefore, a unified model might be considered to fully make use of their advantages and provide a more promising method in tag-aware recommender systems. Nowadays, RS is not a new problem in information science, the advent of new Web2.0 paradigms bring versatile tools and information to help build better recommendation models by integrating traditional methods. Recently, the studies of complex networks would benefit tag-based algorithms, because the in-depth understanding of network structure, user behaviors and network dynamics can be used to design advanced tag-aware recommendation algorithms (e.g., making use of the information about hypergraph BIB016 and tripartite graph BIB005 BIB008 of social tagging networks to better predict underlying interests). On the other hand, tag-based algorithms can also help the trend detection BIB007 over time. Although the studies of tag-aware recommender systems have achieved fruitful goals, there are still challenges, as well as some new directions remained to be solved (discovered) in future: (i) the complete hypergraph BIB013 should be well considered to fully address the integrity of tagging networks without decomposing any information and thus is a promising way to provide recommendations with better performance; (ii) most of current related researches emphasize on recommending single type of nodes, however, predicting the joint node pairs (e.g. item-tag pair BIB017 ) comparatively lacks of study. The joint pair recommendation would provide more personalized preference, hence be a new application of tag-aware recommender systems; (iii) since the tags are freely assigned by users, which consequently results in much noise of added tags. Tag clustering BIB010 BIB014 methods and anti-spam BIB011 technique would be both promising ways to reduce the noise and help provide high-quality recommendations; (iv) the probability-based models are mainly used to provide tag recommendations in most recent researches, while how to well use them to benefit item recommendations is still an open challenge. In addition, those models would also help to prevent rumor spreading BIB001 BIB002 and trend detection BIB004 ; (v) the multi-layered network BIB003 consists of user social interactions, tag co-occurrence relations and user-item-tag ternary information can be considered to describe the hierarchical structure of social tagging systems, and thus the Social Network Analysis (SNA) and social influence BIB015 BIB018 BIB019 based techniques can be used to provide more substantial recommendations, and social predictions BIB012 BIB020 as well; (vi) most tagging platforms are dynamical systems and evolve over time BIB009 BIB023 , thus the study of human dynamics BIB006 in analyzing the temporal behaviors and interests can provide real-time recommendations BIB021 BIB022 .

Agent-based distributed manufacturing control: A state-of-the-art survey <s> ARTICLE IN PRESS <s> The field of holonic manufacturing was initiated in the early 1990s to address the upcoming challenges for manufacturing operations in the 21st century. It is intended to support highly responsive organizations by providing a modular building block or 'plug and play' capability for developing and operating a manufacturing production system. The holonic approach can be viewed as an alternative to more hierarchical operations management methods such as those based on computer- integrated manufacturing (CIM). Since 1990, an increasing amount of research has been conducted in holonic manufacturing over a diverse range of industries and applications, with a strong emphasis on how holonic systems will perform the different planning and control functions required to manage a production operation. The planning and control work to date has, however, been focussed on specific problem formulations and solution strategies. The intention of this paper is to provide an overview of the use of holonic manufacturing conce... <s> BIB001 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> ARTICLE IN PRESS <s> The emerging paradigm of agent-based computation has revolutionized the building of intelligent and decentralized systems. The new technologies met well the requirements in all domains of manufacturing where problems of uncertainty and temporal dynamics, information sharing and distributed operation, or coordination and cooperation of autonomous entities had to be tackled. In the paper software agents and multi-agent systems are introduced and through a comprehensive survey, their potential manufacturing applications are outlined. Special emphasis is laid on methodological issues and deployed industrial systems. After discussing open issues and strategic research directions, we conclude that the evolution of agent technologies and manufacturing will probably proceed hand in hand. The former can receive real challenges from the latter, which, in turn, will have more and more benefits in applying agent technologies, presumably together with well-established or emerging approaches of other disciplines. <s> BIB002 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> ARTICLE IN PRESS <s> This paper surveys the literature in the holonic manufacturing systems area in an attempt to bring together the key issues in the development and applications of holonic systems. A brief introduction presents the characteristics of today’s manufacturing environment and the requirements for next generation manufacturing systems. Then, starting with considerations about the origins of the holonic concept and its first applications in manufacturing, the paper presents the advances made in applying the holonic concept to manufacturing systems area. Several considerations for the development of holonic manufacturing systems and specific holonic system requirements are discussed. As holonic concept is considered a solution for next generation manufacturing systems, there is a significant number of applications and implementations of the holonic concept in manufacturing systems domain. The most important and relevant approaches developed so far are presented. Finally, a short conclusion and future research directions in the area are provided <s> BIB003 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> ARTICLE IN PRESS <s> Agent technology has been considered as an important approach for developing distributed intelligent manufacturing systems. A number of researchers have attempted to apply agent technology to manufacturing enterprise integration, supply chain management, manufacturing planning, scheduling and control, materials handling, and holonic manufacturing systems. This paper gives a brief survey of some related projects in this area, and discusses some key issues in developing agent-based manufacturing systems such as agent technology for enterprise integration and supply chain management, agent encapsulation, system architectures, dynamic system reconfiguration, learning, design and manufacturability assessments, distributed dynamic scheduling, integration of planning and scheduling, concurrent scheduling and execution, factory control structures, potential tools and standards for developing agent-based manufacturing systems. An extensive annotated bibliography is provided. <s> BIB004

This paper presents the state-of-the-art in intelligent and distributed manufacturing control systems using emerging paradigms, such as multi-agent systems and holonic manufacturing systems (HMSs), and surveys the applications of agent-based manufacturing control systems, including the real implementations in industry. The objective of this paper is not to provide an exhaustive survey of the application of multi-agent systems and holonic manufacturing principles to manufacturing environment, but to focus on the manufacturing control applications. Earlier surveys of multi-agent systems for intelligent manufacturing systems (IMSs) can be found in BIB004 , BIB002 , and for the development of holonic manufacturing applications can be found in BIB003 , and BIB001 . This paper also tries to understand why these distributed and intelligent control solutions are not currently more adopted by industry. For this purpose, the possible reasons are identified and discussed, being the challenges and trends in agent-based manufacturing control systems pointed out to increase their performance and adoption. The paper is organized as follows. Section 2 reviews the concepts associated with manufacturing control systems, describing the traditional approaches and the distributed and intelligent ones, namely the agent-based and holonic manufacturing control systems, which address agility, flexibility and adaptability requirements. Section 3 surveys the existing applications of multiagent systems and HMSs in manufacturing control, and Section 4 identifies the reasons for the weak adoption of these emergent control solutions by industry. Section 5 points out the trends, challenges and research opportunities in agent-based agile manufacturing control systems. Finally, Section 6 rounds up the paper with conclusions.

Agent-based distributed manufacturing control: A state-of-the-art survey <s> Traditional approach to manufacturing control problem <s> Part 1 Overview: A background to shop floor control systems the just in time approach to production management an overview of requirements planning (MRP and MRP 2) guidelines for the development and installation of production planning and control systems an architecture for production planning and control. Part 2 A functional architecture for shop floor control systems: an architecture for shop floor control systems a structured functional model for shop floor control. Part 3 An information technology architecture for shop floor control: overview technology implementation technologies for shop floor control systems. Part 4 State-of-the-art review: a review of scheduling strategies a review of production environment design strategies. Part 5 The implementation of shop floor control systems: overview an approach to the implementation of factory co-ordination and production activity control systems a design tool for shop floor control systems. Part 6 An implementation of a PAC system: overview the environment of the case study implementation of a PAC system. <s> BIB001 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> Traditional approach to manufacturing control problem <s> This paper proposes a methodology for the development of collaborative (agent-based) production systems, using high-level Petri nets. The proposed methodology supports the development life-cycle from specifications analysis through to design-validation and implementation of collaborative and re-configurable production systems and their control systems, in an integrated manner. It covers a wide spectrum of application domains, ranging from intelligent mechatronic devices to multi-agent distributed manufacturing control systems. <s> BIB002

The manufacturing control is concerned with managing and controlling the physical activities in the factory aiming to execute the manufacturing plans, provided by the manufacturing planning activity, and to monitor the progress of the product as it is being processed, assembled, moved, and inspected in the factory. Algorithms at this level are used to decide what to produce, how much to produce, when production is to be finished, how and when to use the resources or make them available, when to release jobs into the factory, which jobs to release, job routing, and job/operation sequencing . Due to its complexity, especially the high number of interactions between the different components and the variety of functions executed, manufacturing control systems are traditionally implemented using centralized or hierarchical control approaches, comprising, as illustrated in Fig. 1 , the following main components: planning, scheduling, execution (i.e. dispatching, monitoring, diagnosis and error recovery) and machine/ device control. Each one of these components operates in a specific temporal horizon, ranging from weeks at the strategic level to seconds at the shop floor. The production plans are passed to the scheduling component by the production planning, which have a temporal horizon of days and weeks. The scheduling is concerned with the assignment of operation to resources, within a shorter temporal horizon and respecting a specific criterion, e.g. the due date or priority. Scheduling can be defined as the optimal allocation of resources over the time to jobs, where these assignments must obey to a set of constraints that reflect the temporal relationships between jobs and the capacity limitations of the resources. The manufacturing scheduling is a complex combinatorial problem, more specifically a non-polynomial (NP) problem, widely studied and reported in the literature, mainly due to its highly combinatorial aspects, its dynamic nature and its applicability in manufacturing systems . The scheduling methods range from simple heuristics, such as earliest due date (EDD) or shortest processing time (SPT), to more elaborated computational techniques, such as constraint satisfaction techniques, neighborhood search techniques and genetic algorithms. The execution is related to performing the final assignment of orders to the resources based on the current state of the manufacturing system and on the schedule plan. A dispatching algorithm decides how to use a manufacturing resource only upon the availability of the factory plant resources, taking into account the current status of the production system BIB001 . The dispatching rule determines which job a resource will work on next. This sequencing decision can be based on the job's due date, the customer priority, minimization of set-ups, the SPT, or any other possible rule or heuristic. An important aspect of the factory operation is to have a detailed and up-to-minute knowledge about the work in progress and the status of the process. The execution of manufacturing plans is subjected to deviations, e.g. due to machine failures, operators' absenteeism, rush orders or parts delayed by suppliers, which implies the decrease of the system productivity. In this case, the system should respond dynamically and quickly to the disturbance, taking proper corrective actions to complete the production orders on time and to minimize the impact of the disturbances, for example by reformulating the plans or executing corrective maintenance. The monitoring component also provides information about the progress of the plans execution to the upper-level components, which can, if necessary, reformulate their plans. In case of error detection it is necessary to perform some kind of diagnostics that help to identify the source of malfunctions and unreliable operation conditions. The treatment of disturbances during the execution of the production plans makes the manufacturing control interesting and complex. Machine/device control is the lowest level of control hierarchy and involves the initiation, coordination and monitoring of the different machine functions. The traditional approach to manufacturing control systems based on centralized or hierarchical control structures, presents good characteristics in terms of productivity, essentially due to its intrinsic optimization capabilities. However, dynamic and adaptive response to change is, currently, the key to competitiveness, and the traditional approaches to manufacturing control typically fall into large monolithic and centralized software packages that are developed and adapted case by case, requiring a huge and expensive effort to implement, maintain or re-configure the control application. In conclusion, they are not adequate because they do not support efficiently the current requirements imposed to manufacturing systems, namely in terms of flexibility, expansibility, agility and re-configurability. A new class of intelligent and distributed manufacturing control systems is then required to fulfill the gap left by the centralized approaches, in which BIB002 : Using a distributed approach, a complex problem can be divided into several small problems, each one mapped on an intelligent building block, i.e. control unit; Each control unit is autonomous having its own objectives, knowledge and skills, and encapsulating intelligent functions; however, none of them has a global view of the system; The global control decisions (e.g. the scheduling, monitoring and diagnosis) are determined by more than one control unit, i.e. the control units need to work together, interacting in a collaborative way to reach a production decision; Some control units are connected to physical automation devices, such as robots and CNC machines; Control units should exhibit several important features such as re-configurability, robustness, plugability, learning and reusability. A manufacturing control system that satisfies the above requirements operates in a totally different way when compared with the traditional centralized control systems. The change from the traditional centralized approach to the new distributed and intelligent approach is illustrated in Fig. 2 MESA International, 1995 and distributed and intelligent manufacturing control. These paradigms, introducing artificial intelligence techniques in practice, have the capability to respond promptly and correctly to change, and differ from the conventional approaches due to their inherent capabilities to adapt to emergence without external intervention.

Agent-based distributed manufacturing control: A state-of-the-art survey <s> Agent-based manufacturing control <s> The concept of an agent has become important in both Artificial Intelligence (AI) and mainstream computer science. Our aim in this paper is to point the reader at what we perceive to be the most important theoretical and practical issues associated with the design and construction of intelligent agents. For convenience, we divide these issues into three areas (though as the reader will see, the divisions are at times somewhat arbitrary). Agent theory is concerned with the question of what an agent is, and the use of mathematical formalisms for representing and reasoning about the properties of agents. Agent architectures can be thought of as software engineering models of agents;researchers in this area are primarily concerned with the problem of designing software or hardware systems that will satisfy the properties specified by agent theorists. Finally, agent languages are software systems for programming and experimenting with agents; these languages may embody principles proposed by theorists. The paper is not intended to serve as a tutorial introduction to all the issues mentioned; we hope instead simply to identify the most important issues, and point to work that elaborates on them. The article includes a short review of current and potential applications of agent technology. <s> BIB001 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> Agent-based manufacturing control <s> Agent software is a rapidly developing area of research. However, the overuse of the word ‘agent’ has tended to mask the fact that, in reality, there is a truly heterogeneous body of research being carried out under this banner. This overview paper presents a typology of agents. Next, it places agents in context, defines them and then goes on, inter alia, to overview critically the rationales, hypotheses, goals, challenges and state-of-the-art demonstrators of the various agent types in our typology. Hence, it attempts to make explicit much of what is usually implicit in the agents literature. It also proceeds to overview some other general issues which pertain to all the types of agents in the typology. This paper largely reviews software agents, and it also contains some strong opinions that are not necessarily widely accepted by the agent community. <s> BIB002 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> Agent-based manufacturing control <s> The Center for Electronic Commerce (CEC) embodies over fourteen years of experience in applying agents to industrial problems. We have found such a fit in three areas: coordination of industrial designers, simulation and modeling of complex products and processes, and scheduling and control of production systems. This presentation outlines several trends in modern manufacturing, describes how these trends affect the three problem areas, discusses the features of agents that make them attractive candidates for implementing such systems, and reviews example applications from CEC's portfolio in each of these areas. <s> BIB003 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> Agent-based manufacturing control <s> From the Publisher: ::: What are multi-agent systems? How do they work? What do they do? If you are looking for the answers to these questions, read on; for Jacques Ferber's authoritative book is the first to provide a single, coherent overview of multi-agent systems. Introduces and defines key concepts throughout the text; provides numerous examples to illustrate core principles; draws on contributions from different disciplines to present a holistic, comprehensive picture of state-of-the art agent technology; and describes all the latest developments in the field and encourages the reader to reflect on possibilities for the future. <s> BIB004

The multi-agent system paradigm derives from the distributed artificial intelligence (DAI) field, being characterized by decentralization and parallel execution of activities based on autonomous entities, called agents. The definition of agent concept is neither unique nor consensual BIB001 BIB004 . Despite some definitions and interpretations for agents, a suitable definition is: ''An autonomous component that represents physical or logical objects in the system, capable to act in order to achieve its goals, and being able to interact with other agents, when it does not possess knowledge and skills to reach alone its objectives''. The most important properties of an agent are the autonomy, intelligence, adaptation and co-operation. There are several agent architectures, ranging from reactive agents, operating in a stimulus-response manner, to deliberative agents characterized by their pro-active reasoning and goaloriented behavior. A well-known deliberative and cognitive agenttype is belief-desire-intention (BDI) architecture, which origin lies in a theory of human practical reasoning, focusing particularly on the role of intentions in practical reasoning . In the BDI agents, the decision-making depends on the manipulation of beliefs, desires and intentions of the agents. The development of reactive agents is simpler than the cognitive agents BIB004 , easier to understand and more robust and fault-tolerant than the other agent types BIB002 . However, reactive agents are incapable of foreseeing what is going to happen and thus of anticipating the future by planning what action to take BIB004 . A multi-agent system can be defined as a set of agents that represent the objects of a system, capable of interacting, in order to achieve their individual goals, when they have not enough knowledge and/or skills to achieve individually their objectives. Agents organize themselves into a heterarchical structure characterized by the high-level of autonomy and co-operation, being the client-server structure with fixed relations no more applied . These features allow a high performance against disturbances, being the global optimization reduced, because the decision-making is local and autonomous, without a global view of the system. The expansibility of the system is easier, being only enough to modify the functioning of some agents or add new agents to the control system. Multi-agents systems allow a new approach to the problems, both in the design and implementation phases, introducing functionalities that support efficiently the distributed manufacturing system needs, such as modularity, decentralization and dynamic and complex structures characteristics, for what agents are well suited to solve BIB003 . Additionally, the required software to develop agents is shorter and simpler than the software required by the centralized approach, leading to easier development, debug and maintenance . In multi-agent systems, since each agent has a partial view of the system, the agents need to be able to communicate in order to achieve a pre-defined goal or solve a problem. The interaction between agents requires that the agents can understand themselves, using a proper agent communication language, ontologies and interaction protocols. In volatile and dynamic scenarios, where it is difficult to foresee future events, agents must learn to adapt their behavior to those dynamic environments, improving their performance. Learning capabilities contribute to the intelligence of an agent, by acquiring new knowledge and skills, which will be used in the future to take better decisions. Agent-based approaches have been applied in many different areas, such as electronic commerce, e-business, air traffic control, process control and telecommunications, besides manufacturing. In fact, manufacturing, transport, telecommunications and healthcare are seen as the most significant domains for agent technology . In the automation and manufacturing domains, an agent can represent physical resources, such as machine tools, robots, auto-guided vehicles (AGVs) and products and logical objects, such as the schedulers and orders. Using the appropriate distributed control algorithms, individual machines and product agents can make their own manufacturing control decisions relating to resource allocation and coordination, using an automated form of ''negotiation''. The key benefit of such approach is that if production is disrupted or re-organized in some way, the same negotiation process still takes place, albeit with different machines or products making the decisions, and hence the system is relatively robust to change. Fig. 3 illustrates, with a simple example, how the production control works using agents. This example comprises: (i) a part agent, running in an industrial personal computer (IPC), that is responsible for the supervision of the execution of operations in the part, (ii) three machine agents representing three computer numerical control (CNC) machines available at shop floor, running in the numerical controller of each machine and (iii) a transport agent representing an AGV and running in a programmable logic controller (PLC) and programmed using IEC61131-3 language. Initially, the part agent queries the available agents that represent the factory resources, about who has skills and is available to execute a drilling operation in the part. Independently, each agent verifies its skills and availability, and answer to the part agent: The agent representing machine #1 replies the part agent saying that it cannot execute the operation since it is out of service. The agent representing machine #2 replies the part agent saying that it is overloaded. The agent representing machine #3 answers positively to the announcement. At the end of the negotiation, the part agent allocates the drilling operation to the machine #3, and then negotiates with the transporter agent(s), the transportation of the part to the physical location of machine #3. It is important to notice that the control system, in the illustrated example, is independent of the number of machines in the system, as well it does not ''feel'' the introduction of new machines or the remotion of machines. Additionally, the agents representing the several machines were developed using the same ''piece of software'', being customized for each machine according to its type, skills and behavior.

Agent-based distributed manufacturing control: A state-of-the-art survey <s> Holonic manufacturing control <s> Koestler examines the notion that the parts of the human brain-structure which account for reason and emotion are not fully coordinated. This kind of deficiency may explain the paranoia, violence, and insanity that are central parts of human history, according to Koestler's challenging analysis of the human predicament. <s> BIB001 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> Holonic manufacturing control <s> 430,091. Insulating cable joints by extrusion - machines. ELECTRICAL RESEARCH PRODUCTS, Inc., 195, Broadway, New York, U.S.A.-(Assignees of Malm, F. S. ; 936, Ridgewood Road, Millburn, New York, U.S.A.) Dec. 13, 1933, No. 35130. Convention date, Dec. 29,1932. [Class 87 (ii)] [See also Group XXXVI] For making good the insulation after the jointing of cables, the cable is inserted in clamps 23 and stretched by separating the standard 20 from the fixed standard 19 by the screw-mechanism 21. The supports are then moved laterally on the guides 12 until the cable enters the open side 35, Fig. 2, of an extrusion head, with the joint well clear of the left hand side of the head as viewed in Fig. 1. Perforated tapered dies in mating halves are then placed on the cable between the head and the joint and the cable is then fed axially through the head by rotation of the screw 13 either by hand from the crank 14 or by power from the clutch 15 engaged by operating the handle 18. The dies travel with the cable until they wedge in the head, after which the continued movement of the cable draws the joint through the dies. Insulating strip material from the reservoir 24 is masticated by the rollers 25 and then delivered by the tapered feed screw 27 through the tapering valved outlet 28 leading to the extrusion head, thence passing through the orifices in the dies to the cable. Byepass conduits 33 may be opened to allow some of the material to escape to the exterior of the machine in order to equalize the pressure at opposite sides of the dies and to fill all the parts of the machine at starting, before making a joint. The extrusion head is of massive construction with walls approximately equal in thickness to the radius of the die orifice. The casing of the feed screw is electrically heated, and the size of the outlet die is controlled by the sliding members 29 operated by screw-mechanism 30 working in conjunction with a pointer 31 and scale. The subject-matter of Specification 389,874 is disclaimed. <s> BIB002 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> Holonic manufacturing control <s> This contribution introduces in the background of holonic systems and holarchies as a hierachical model for biological and social structures in nature. The characteristics of these hierarchical architectures and its elements will be explained. It will be shown how this approach can be used as a potential solution for present and future challenges in manufacturing control systems. The results of an internationally performed feasability study heading for the joint development of this technology will be described. A report on the current status of the international research activities will cover the technical and organisational aspects of the project. <s> BIB003 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> Holonic manufacturing control <s> Abstract Future manufacturing systems need to cope with frequent changes and disturbances. As such, their control requires constant adaptation and high flexibility. Holonic manufacturing is a highly distributed control paradigm that promises to handle these problems successfully. It is based on the concept of autonomous co-operating agents, called `holons'. This paper gives an overview of the holonic reference architecture for manufacturing systems as developed at PMA-KULeuven. This architecture, called PROSA, consists of three types of basic holons: order holons, product holons, and resource holons. They are structured using the object-oriented concepts of aggregation and specialisation. Staff holons can be added to assist the basic holons with expert knowledge. The resulting architecture has a high degree of self-similarity, which reduces the complexity to integrate new components and enables easy reconfiguration of the system. PROSA shows to cover aspects of both hierarchical as well as heterarchical control approaches. As such, it can be regarded as a generalisation of the two former approaches. More importantly, PROSA introduces significant innovations: the system structure is decoupled from the control algorithm, logistical aspects can be decoupled from technical ones, and PROSA opens opportunities to achieve more advanced hybrid control algorithms. <s> BIB004

To face the requirements of operating on a global scale and to meet the needs of an ever more demanding consumer market, an international collaborative research program in manufacturing, called IMS, was started in the beginning of nineties. Within the IMS programme, several paradigms for the factory of the future were developed, such as holonic, bionic and fractal manufacturing systems. These theories present similar concepts and characteristics but with different origins: mathematics for the fractal factory , nature for bionic manufacturing systems and social organizations for HMSs BIB004 . These paradigms suggest the idea that manufacturing systems will continue to need a hierarchical structure besides the increased autonomy assigned to individual entities. They also advise that hierarchy is needed in order to guarantee the interentities conflict resolution and to maintain the overall system coherence and objectivity resulting from the individual and autonomous attitude of the entities . In spite of the similarity of concepts and characteristics, these paradigms emphasize a different set of issues and characteristics, as reviewed by . The HMS is a paradigm that translates into the manufacturing world the concepts developed by Arthur Koestler to living organisms and social organizations. In middle of sixties, Koestler introduced the word holon to describe the basic unit of organization in living organisms and social organizations, based on Herbert Simon theories and on his observations BIB001 . Simon observed that complex systems are hierarchical systems formed by intermediate stable forms, which do not exist as auto-sufficient and non-interactive elements but, on the contrary, they are simultaneously a part and a whole. Koestler concluded that parts and wholes do not exist in domain of life, and proposed the word holon to represent this hybrid nature, being a combination of the Greek word holos, which means whole, and the suffix on, which means particle. BIB001 also identified two important characteristics of a holon: Autonomy, where the stability of the holons result from their ability to act autonomously in case of unpredictable circumstances; Co-operation, which is the ability to have holons cooperating, transforming these holons into effective components of bigger wholes. A holon can represent a physical or logical activity, such as a robot, a machine, an order, a flexible manufacturing system or even an human operator. The holon has information about itself and the environment, containing an information processing part and a physical processing part when the holon represents a physical device BIB003 , such as an industrial robot, as illustrated in Fig. 4 . A holarchy is defined as a system of holons, organized in a hierarchical structure, cooperating to achieve the system goals, by combining their individual skills and knowledge. Each holarchy has fixed rules and directives, and a holon can dynamically belong to multiple holarchies at the same time, which is an important difference to the traditional concept of hierarchies. The holons can integrate themselves into a holarchy and, at the same time, to preserve their autonomy and individuality. A HMS is a holarchy that integrates the entire range of manufacturing activities from order booking through design, production and marketing to realize the agile manufacturing enterprise BIB002 . In HMS, the holon's behaviors and activities are determined through co-operation with other holons, as opposed to being determined by a centralized mechanism. Applying these concepts, HMS can be used to implement control structures that combine the advantages of hierarchical and heterarchical approaches, showing the reactivity against disturbances presented in heterarchical control, and the high and predictable performance presented in hierarchical control . According to the Janus effect, i.e. that holons combine the whole and the part, being simultaneously self-contained wholes to their subordinated parts and dependent parts when seen from higher levels, it is possible to decompose a holon into several others holons, which in turn can be broken into further holons, allowing the reduction of the problem complexity. This feature allows the structural development of production control systems through the encapsulation of manufacturing functions and components. The implementation of the holonic manufacturing concepts can be done using the agent technology, which is appropriate to implement modularity, decentralization, reuse and complex structures characteristics. The use of agent technology addresses mainly the high-level of abstraction , as illustrated in Fig. 5 . At the lowest real-time control level, the interconnection with physical devices is required, making it able to read data from sensors and to send actions to actuators. Currently, the lowest real-time control is usually carried out by industrial PLCs running in a classical scan-based manner that ensures the real-time responsiveness of the control system and provides natural I/O connectivity to the real manufacturing process. The control programs here are developed using IEC 61131-3 standard programming languages, particularly the ladder logic. A possible alternative is to use the IEC 61499 function blocks standard that is an extension of the IEC 61131-3 function block diagrams. IEC 61499 defines a new way to model the control and execution of algorithms in distributed control systems, by encapsulating and reusing software modules ).

Agent-based distributed manufacturing control: A state-of-the-art survey <s> Architectures and approaches <s> Holonic architectures have been suggested as a possible building block for future manufacturing systems. The term holonic implies a unit in manufacturing that demonstrates the dual characteristics of autonomous behaviour (when required) and the ability to function cooperatively when the situation requires it. One of the key functions that is important to both of these properties is the need for the manufacturing unit to be able to (self) diagnose performance and functional failures in a systematic manner. This paper proposes a generic methodology for self-diagnosis within a holonic manufacturing system, in which faults and problems are analysed as a standard function of the system. <s> BIB001 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> Architectures and approaches <s> Abstract The concept of holonic manufacturing is based on the cooperation of autonomous, functionally complete entities with diverse, often conflicting goals. The paper introduces a market mechanism for coordinating the activities of intelligent agents that pursue their own interest by operating under bounded rationality in a changing, hardly predictable environment. The market model is used for solving dynamic order processing and scheduling problems: conflicts between local scheduling agents are resolved by negotiating and bargaining on simple common terms of tasks, due dates and prices. <s> BIB002 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> Architectures and approaches <s> A generic Mediator architecture for distributed task planning and coordination has been developed using multi-agent paradigms. In this approach, agents function autonomously as independent computing processes, and dynamic virtual clusters coordinate the agent's activities and decision making. This coordination involves dynamically created coordination agents and resource agents concurrently. The Mediator architecture contains three levels of these coordination agents: the template mediator, the data-agent manager, and the active mediator. The template mediator is the top-level global coordinator. This agent contains both the templates and the cloning mechanism to create the successively lower-level agents. Task plans are decomposed successively into subtasks, which are allocated to dynamically created agent clusters coordinated through data-agent managers and active mediators. Coordination of agent activity takes place both among the clusters and within each cluster. The system dynamically adapts to evolving manufacturing tasks, with virtual agent clusters being created as needed, and destroyed when their tasks are completed. The mediator architecture and related mechanisms are demonstrated using an intelligent manufacturing scheduling application. Both the machines and the parts involved in this production system are considered as intelligent agents. These agents use a common language protocol based on the Knowledge Query Manipulation Language (KQML). The generic Mediator approach can be used for other distributed organizational systems beyond the intelligent manufacturing application it was originally developed for. <s> BIB003 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> Architectures and approaches <s> While current research into decentralized control architectures for advanced manufacturing systems has proven the concept of architectures ranging from hierarchical to heterarchical control, neither extreme of the control architecture spectrum has proven to be the most appropriate choice for given manufacturing systems. This paper reports on research into a control architecture that does show promise in this area: a hybrid control architecture consisting of elements of both hierarchical and heterarchical control architectures. First control architectural issues are considered. Next, the concepts of virtual control organizations and partial dynamic hierarchies are presented. The paper then concludes with a discussion of recent work on implementing partial dynamic hierarchies within real-time control constraints, using an agent-based architecture. <s> BIB004 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> Architectures and approaches <s> Abstract This paper describes ongoing research and development of a Holonic Machine Controller. The characteristics of this new prototype controller will be more flexible and open in comparison with present NC technology. PC based software has been developed on a Windows 16-bit platform that embeds the full functionality of the Machine Control Unit (MCU) of a five axis milling machine. The input to the MCU should be a feature based part description. The new controller is intended to autonomously generate machining NC sequences that are stepwise downloaded and executed by the original NC controller, using the LSV-2 communication protocol. The protocol is conceived as a device-driver. A device-driver can process NIOs (Non-Interruptable Operations) for machining, process gauging, etc. Holonic behaviour will be supported by advanced planning, execution, and monitoring (PEM) actions. <s> BIB005 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> Architectures and approaches <s> The Center for Electronic Commerce (CEC) embodies over fourteen years of experience in applying agents to industrial problems. We have found such a fit in three areas: coordination of industrial designers, simulation and modeling of complex products and processes, and scheduling and control of production systems. This presentation outlines several trends in modern manufacturing, describes how these trends affect the three problem areas, discusses the features of agents that make them attractive candidates for implementing such systems, and reviews example applications from CEC's portfolio in each of these areas. <s> BIB006 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> Architectures and approaches <s> Abstract A Holonic Manufacturing System (HMS) is a manufacturing system where key elements, such as machines, cells, factories, parts, products, operators, teams, etc., are modeled as `holons' having autonomous and cooperative properties. The decentralized information structure, the distributed decision-making authority, the integration of physical and informational aspects, and the cooperative relationship among holons, make the HMS a new paradigm, with great potential for meeting today's agile manufacturing challenges. Critical issues to be investigated include how to define holons for a given problem context, what should be the appropriate system architecture, and how to design effective cooperation mechanisms for good system performance. In this paper, holonic scheduling is developed for a factory consisting of multiple cells. Relevant holons are identified, and their relationships are delineated through a novel modeling of the interactions among parts, machines, and cells. The cooperation mechanisms among holons are established based on the pricing concept of market economy following `Lagrangian relaxation' of mathematical optimization, and cooperation across cells is performed without accessing individual cells' local information nor intruding on their decision authority. The system also possesses structural recursivity and extendibility. Numerical testing shows that the method can generate near-optimal schedules with quantifiable quality in a timely fashion, and has comparable computational requirements and performance as compared to the centralized method following single-level Lagrangian relaxation. The method thus provides a theoretical foundation for guiding the cooperation among holons, leading to globally near-optimal performance. <s> BIB007 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> Architectures and approaches <s> Abstract This paper deals with a new architecture and negotiation protocol for the dynamic scheduling of Manufacturing Systems. The architecture is based on two paradigms: Multi-Agent Systems and Holonic Systems. The main contribution in the architecture is the development of Holons representing tasks and resources. The well-known Contract Net Protocol [R. Davis, R. Smith, Negotiation as a metaphor for distributed problem solving, Artificial Intelligence, 20, (1), (1983) 63–109] has been adapted to handle temporal constraints and to deal with scheduling conflicts. The purpose of this protocol is to dynamically assign operations to the resources of the Manufacturing System to accomplish the proposed tasks. This protocol involves a renegotiation phase whenever exceptions appear. It also deals with conflict situations, namely, with the case of the `indecision problem'. The used approach assumes that deadlines are the most important constraints to consider. Thus the acceptance or refusal of a resource for a specific operation depends on the capability of executing the operation within the specified deadline. <s> BIB008 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> Architectures and approaches <s> Abstract This article presents a new approach to the design of the architecture of a computer-integrated manufacturing (CIM) system. It starts with presenting the basic ideas of novel approaches which are best characterised as fractal or holonic models for the design of flexible manufacturing systems (FMS). The article discusses hierarchical and decentralised concepts for the design of such systems and argues that software agents are the ideal means for their implementation. The agent architecture InteRRaP for agent design is presented and is used to describe a planning and control architecture for a CIM system, which is separated into the layer of the production planning and control system, the shop floor control systems, the flexible cell control layer, the autonomous system layer, and the machine control layer. Two application scenarios are described at the end of the article and results are reported which were obtained from experiments with the implementations for these application scenarios. While one of these scenarios — a model of an FMS — is more research-oriented, the second one — optimisation of a production line — is directly related to an industrial real-world setting. <s> BIB009 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> Architectures and approaches <s> Management of complexity, changes and disturbances is one of the key issues in production today. Distributed, agent-based (holonic) structures represent viable alternatives to hierachical systems provided with reactive/proactive capabilities. The paper outlines the difficulties which hinder their industrial acceptance. Several approaches to overcome these barriers are introduced, i.e. the use of simulation techniques for developing and testing agent-based control architectures, the holonification of existing resources and traditional (centralized/hierarchical) manufacturing systems. Finally, the cooperative use of agent-based distributed control structures and the more centralized (e.g. GA-based) schedulers is proposed aiming at systems which can handle critical complexity, reactivity, disturbance and optimality issues at the same time. <s> BIB010 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> Architectures and approaches <s> As the existing control architectures for FMS have many problems with respect to flexibility, re-configuration and software development, which become worse as the system size increases, the international IMS research programme proposed a new manufacturing philosophy called the Holonic Manufacturing System (HMS). Under the notion of HMS, manufacturing systems are built from a modular mix of standardised, autonomous and co-operative agents called holons. Inspired by such emerging concept, we carried out this research with the aim of devising a holonic control framework, which is suitable for future FMS shop floor scheduling and control. This paper addresses the design of the proposed framework as well as the underlying holonic philosophy. In order to verify the applicability of the framework, an experimental simulation system has been developed on a real-time and distributed computer platform. The system consists of a set of holon prototypes capable of performing real-time scheduling for both manufacturing and transportation tasks, and a collection of interacting software modules that assist the simulation running on these prototypes. Experimental results acquired from the simulation have demonstrated that the framework is extensible and re-configurable. Implementation of the simulation system also provides direction and a workable foundation for further research and development on HMS. <s> BIB011 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> Architectures and approaches <s> Automated guided vehicle (AGV) systems traditionally utilize a central controller to assign transportation tasks among vehicles. Because of its rigidity, a centralized architecture may not suitable for modern manufacturing with its rapidly changing markets, demands for products with higher quality, smaller batch sizes, customer specifications, and shorter delivery times. This research applies concepts from holonic manufacturing systems to develop a distributed AGV control system that is flexible and robust in the presence of disturbances. We present the architecture of the holonic manufacturing system, a coordination strategy based on contract nets, and simulation results that indicate an improvement in utilization when compared with traditional centralized control. <s> BIB012 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> Architectures and approaches <s> Holonic manufacturing systems have emerged over the last seven years (1993-2000) as strategy for manufacturing control system design. A new approach called Holonic Component-Based Architecture (HCBA) to establish a manufacturing control system as to cope with rapid changes in manufacturing environment is presented. Intelligent building blocks in terms of resource and product are proposed to dynamically form a virtual controller via a computer network, and to perform co-operative control execution and diagnosis operations. This concept enables the design, operation and maintenance of the manufacturing controller to be performed in a distributed manner, which can increase the agility and responsiveness of an integrated system. This flexible structure has been implemented in a robot assembly cell to show its plug-and-play capability via an Internet based infrastructure. <s> BIB013 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> Architectures and approaches <s> Abstract The paper proposes a novel approach to the design of line-less production systems. The approach is based on the concept of biological manufacturing systems (BMS) that can deal with complexity in manufacturing using the ideas of self-organization, evolution and learning. In the line-less production systems, all production elements can move freely on the production floor using self-organization in order to adapt to fluctuations such as the diversity of production demands and the malfunction of machines. The effectiveness of the line-less production system is discussed and its feasibility is demonstrated by computer simulation. Furthermore, a mini-factory consisting of small autonomous robots is developed as a prototype of the line-less production systems. <s> BIB014 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> Architectures and approaches <s> We describe a general approach for dynamic and intelligent reconfiguration of real-time distributed control systems that utilizes the IEC 61499 function block model. This work is central to the development of distributed intelligent control systems that are inherently adaptable and dynamically reconfigurable. The approach that is used takes advantage of distributed artificial intelligence at the planning and control levels to achieve significantly shorter up-front commissioning times as well as significantly more responsiveness to change. This approach is based on object-oriented and agent-based methods, and aims at overcoming the difficulties associated with managing real-time reconfiguration of an intelligent manufacturing system. <s> BIB015 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> Architectures and approaches <s> In this paper, we present the system architecture of a flexible manufacturing planning and control system, named EtoPlan. The concept is based on the holonic control approach of building multiple and temporary hierarchies (holarchies). This paper describes the system architecture for flexible planning and control of activities and (groups of) Resources in a manufacture-to-order environment. The system architecture consists of generic control modules that can be applied on different hierarchical levels and for different kinds of manufacturing activities. The main function of the Resource Controller is the Determine Applicability function. <s> BIB016 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> Architectures and approaches <s> Existing modeling frameworks for manufacturing system control can be classified into hierarchical, heterarchical, and hybrid control frameworks. The main drawbacks of existing frameworks are discussed in this paper. A new hybrid modeling framework is also described. It is a hybrid of the two: hierarchical and heterarchical frameworks. In this proposed framework, entities (e.g., parts) and resources (e.g., material handling devices, machines, cells, departments) are modeled as holonic structures that use intelligent agents to function in a cooperative manner so as to accomplish individual, as well as cell-wide and system-wide objectives. To overcome the structural rigidity and lack of flexibility, negotiation mechanisms for real-time task allocation are used. Lower-level holons may autonomously make their negotiations within the boundary conditions that the higher-level holons set. Horizontal, as well as vertical decisions, are made between various levels of controllers, and these are explicitly captured in the model. <s> BIB017 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> Architectures and approaches <s> Abstract Future manufacturing systems need to cope with frequent changes and disturbances. As such, their control requires constant adaptation and high flexibility. Holonic manufacturing is a highly distributed control paradigm that promises to handle these problems successfully. It is based on the concept of autonomous co-operating agents, called `holons'. This paper gives an overview of the holonic reference architecture for manufacturing systems as developed at PMA-KULeuven. This architecture, called PROSA, consists of three types of basic holons: order holons, product holons, and resource holons. They are structured using the object-oriented concepts of aggregation and specialisation. Staff holons can be added to assist the basic holons with expert knowledge. The resulting architecture has a high degree of self-similarity, which reduces the complexity to integrate new components and enables easy reconfiguration of the system. PROSA shows to cover aspects of both hierarchical as well as heterarchical control approaches. As such, it can be regarded as a generalisation of the two former approaches. More importantly, PROSA introduces significant innovations: the system structure is decoupled from the control algorithm, logistical aspects can be decoupled from technical ones, and PROSA opens opportunities to achieve more advanced hybrid control algorithms. <s> BIB018 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> Architectures and approaches <s> In this paper, we present results on designing the architecture of an agent-based system for production control of semiconductor wafer fabrication facilities (wafer fabs). These manufacturing systems are characterized by reentrant product flows, sequence dependant setup-times, inhomogenous parallel machines, a diverse product mix, a mix of different process types, prescribed due dates of the orders and preventive maintenance due to difficult technological processes. Hence, coordination issues have to be considered during the design of a production control system for such type of manufacturing processes. We suggest a hierarchical production control scheme. We use the PROSA reference architecture in order to form the proper agency. We describe the development of a system prototype using the C++ and C# programming language and the.NET middleware. In order to allow the investigation of dynamic scenarios by emulation of the production process we suggest the use of the discrete event simulator AutoSched AP and a blackboard type data layer as a coupling component between the production control system and the simulator. <s> BIB019 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> Architectures and approaches <s> In this article, an agent-based architecture in which cooperation is regulated by contracts is proposed as a flexible approach to dynamic shop floor re-engineering. It describes the dynamic and flexible cooperation of manufacturing agents, representing manufacturing resources and how they can be created from a generic agent template. Agents' behaviour and contract types and structure are discussed. The first experimental results are also introduced. <s> BIB020 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> Architectures and approaches <s> In today's global economy, manufacturing companies must be able to change both the product variety and production quantity without incurring major disturbances in the production process. A manufacturing system capable of performing these changes efficiently needs a control system flexible enough to go from one state to another without significant delays in production. To compensate for the deficiencies of both hierarchical and heterarchical control systems, a control architecture based on the holonic concept is proposed. After presenting the general holonic control architecture, the paper focuses on the integration of the holonic-based control concept in the design of an automated material-handling control system. When performing transport activities in a manufacturing system, due to its architecture, the holonic system can operate both as a hierarchical system, following a predefined schedule during normal operation conditions, and as a heterarchical system in the presence of disturbances. <s> BIB021 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> Architectures and approaches <s> A connector assembly for an optical cable, consisting of an optical fiber and a protective jacket, includes a connector body with an opening therethrough. The opening consists of two coaxial bores of different sizes. A sleeve assembly located within the larger bore includes an outer heat-shrinkable sleeve and a concentric, inner sleeve of meltable filler material. A length of optical fiber from which the protective jacket has been stripped is inserted through the sleeve assembly and through the second, smaller bore in the connector body. Heat is applied to the connector assembly. The shrinking outer sleeve forces melting filler material into the smaller bore to lock the fiber in place. The connector body is also secured to the protective jacket of the optical cable to relieve strain on the optical fiber. <s> BIB022 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> Architectures and approaches <s> Global competition is driving manufacturing companies to change the way they do business. New kinds of shop floor control systems need to be implemented for these companies to respond quickly to changing shop floor environments and customer demands. This paper presents a new concept called iShopFloor-an intelligent shop floor based on the Internet, web, and agent technologies. It focuses on the implementation of distributed intelligence in the manufacturing shop floor. The proposed approach provides the framework for components of a complex control system to work together as a whole rather than as a disjoint set. It encompasses both information architecture and integration methodologies. The paper introduces the basic concept of iShopFloor, a generic system architecture, and system components. It also describes the implementation of eXtensible Markup Language message services in iShopFloor and the application of intelligent agents to distributed manufacturing scheduling. A prototype environment is presented, and some implementation issues are discussed. <s> BIB023 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> Architectures and approaches <s> Abstract The paper discusses the role of simulation in agent-based control systems. We present the universal runtime interface enabling efficient interactions between the agent control, manufacturing environment (emulated or physically connected), visualization and low-level real-time control programs. As a case study, we provide the description of the MAST simulation tool and its application to Cambridge packing cell. <s> BIB024 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> Architectures and approaches <s> In the last decades significant changes in the manufacturing environment have been noticed: moving from a local economy towards a global economy, with markets asking for products with higher quality at lower costs, highly customised and with short life cycle. In these circumstances, the challenge is to develop manufacturing control systems with intelligence capabilities, fast adaptation to the environment changes and more robustness against the occurrence of disturbances. This paper presents an agile and adaptive manufacturing control architecture that addresses the need for the fast reaction to disturbances at the shop floor level, increasing the agility and flexibility of the enterprise, when it works in volatile environments. The proposed architecture introduces an adaptive control that balances dynamically between a more centralised structure and a more decentralised one, allowing combining the global production optimisation with agile reaction to unexpected disturbances. <s> BIB025

Yet another manufacturing system (YAMS) BIB006 ) applies a contract net technique to a hierarchical model of manufacturing system, including agents to represent the shop floor. The autonomous agents at Rock Island Arsenal (AARIA) intends to control a production system to fulfill incoming tasks in due time, focusing on the dynamic scheduling, dynamic reconfiguration and in the control of manufacturing systems that fulfill the deliver dates . The manufacturing resources, process and operations are encapsulated as agents using an autonomous agent approach. Holonic or agent-based scheduling differs from the traditional scheduling in terms of the distribution of the computation and decision-making functions, bringing benefits in terms of reaction to disturbances and the parallel computation. Here the dominant interchange mechanism used to support distributed problem solving is the Contract Net Protocol . BIB008 propose a dynamic scheduling system supported by a holonic approach, using forward and backward influence in the negotiation leading to the task allocation, to handle the temporal constraints and to solve conflicts. The architecture is composed by holons to represent resources, tasks, planning systems, etc. BIB007 define a scheduling algorithm based on Lagrange relaxation concepts. It requires a centralized coordination that guides the individual holons to improve their schedule. BIB002 proposed a market model to solve dynamic order processing and scheduling problems, such as conflicts between local scheduling agents, resolved by negotiation simple terms of tasks, due dates and prices. proposed a holonic approach for manufacturing scheduling and control in a manufacturing cell. modelled the manufacturing operations using an object-oriented approach and propose a real-time scheduling mechanism for assembly lines. BIB011 describes a holonic system where series of prototype holons are implemented for real-time scheduling tasks in an existing FMS. The ADDYMS developed a dynamic scheduling mechanism for local resource allocation at the local work-cell level, using agents to represent physical resources. The product-resource-order-staff architecture (PROSA) BIB018 ) is a holonic reference architecture for manufacturing systems, which uses holons to represent products, resources, orders and logical activities. This architecture is based on three types of basic holons: product, order and resource. The resource holon contains the resource and an information processing part that controls the resource. The product holon holds the process and product knowledge, and contains all information about the product. The order holon represents the tasks in manufacturing systems. Additionally, the architecture defines staff holons, whose mission is to assist and advise the basic other holons. Under the MASCADA project, manufacturing control mechanisms were developed to support the production change and disturbance, safeguarding and/or maximizing the production systems throughput, where the disturbances reduce the effectiveness of the plans/schedules that are generated initially . The approach uses a pro-active disturbance handling mechanism and uses autonomous and intelligent agents to represent the factory components. described the application of a PROSA-based system, developed under the MASCADA project, in a car body painting at the Daimler-Benz plant in Sindelfingen, Germany. BIB012 used the PROSA reference architecture to develop a control architecture for an AGV system capable of being robust in the presence of disturbances. A PROSA-based agent system for production control of semiconductor wafer fabrication facilities entitled FABMAS is reported by BIB019 , where the agents represents work cells, work areas, lots and tools. The discrete event simulator AutoSched AP is used to simulate the behavior of the wafer fabrication shop floor. MetaMorph BIB003 uses an agent federation centered in the mediator approach, supporting the change of form, structure and activity, in order to adapt dynamically to emerging tasks and environment change. Agents represent manufacturing devices and products, and the mediators are used to coordinate the interactions between agents. The approach supports dynamic clustering and cloning, and learning. Tö nshoff and introduced the holonic concepts for the shop floor control and BIB013 presented a specific holonic control system architecture, holonic component-based architecture (HCBA), introducing resource and product holons, to enable a smooth migration between the available standard control hardware and the system needed to implement holonic control. The ADACOR (ADAptive holonic COntrol aRchitecture for distributed manufacturing systems) BIB025 holonic manufacturing control architecture addresses the agile reaction to emergence and change, increasing the agility and flexibility of the enterprise when it works in volatile environments, characterized by the frequent occurrence of disturbances. For this purpose, it introduces an adaptive control approach that evolves in time to combine the global production optimization with the agile reaction to disturbances, being the supervisor entities and the self-organization and learning capabilities associated to the holons, the key roles to support the dynamic evolution and reconfiguration of the organizational control structure. BIB004 used a hybrid control architecture, designated by partial dynamic control hierarchy (PDCH), combining the hierarchical and heterarchical approaches, for manufacturing control. Later, PDCH was used with IEC 61499 function block model to achieve a general approach for dynamic and intelligent reconfiguration of real-time distributed manufacturing control systems BIB015 . The IntaPS project presents an approach for integrated process planning and production control, which architecture consists of two main components, which link information systems of earlier stages of product development and the resources on the shop floor. This link is realized by decentralized planning on shop floor level and by rough level process planning. BIB009 uses a holonic approach for planning and control, built upon the Integration of Reactive behaviour and Rational Planning (InteRRap) hybrid agent architecture , consisting of production planning and control, shop floor control system, flexible cell control, autonomous systems and machine controller levels. The architecture uses agents to represent holonic manufacturing components, forming a multi-agent system organized in a hierarchical structure based on rules. BIB010 proposed a holonification approach of existing resources by incorporating control units for each resource and BIB023 introduced the iShopFloor concept that focuses the implementation of distributed intelligence in the manufacturing shop floor, using intelligent agents, specifically to achieve distributed manufacturing scheduling. BIB016 introduced the Engineer-to-order Planning (EtoPlan) that is a holonic architecture for manufacturing planning and control that aims to deal with large amounts of uncertainty caused by incomplete and unreliable information. The PABADIS ) uses the concept of co-operative manufacturing units (CMUs) to provide significant functions to the production process in automation control, encapsulating residential, products and shop floor management as agents. The approach comprises centralized (for the connection with enterprise resource planning (ERP) systems) and decentralized components, being the products implemented using the mobile agent technology. BIB017 presented a hybrid manufacturing control architecture, i.e. somewhere between the hierarchical and heterarchical control approaches. It defines three levels of agents: higher level with agents elaborating optimal schedules, lower level with agents responding for the individual schedules, and intermediate agents acting as coordinator of actions of lower agents. This holonic schedule and control architecture was applied in an industrial automated warehouse system for order picking and replenishing. BIB021 present a holonic control architecture for automated material handling systems that addresses reliability, real-time scheduling, fault-tolerance and material handling hardware re-configurability. BIB024 described an agent-based simulation environment manufacturing agent simulation tool (MAST) for the transportation of work-pieces among different manufacturing cells using AGVs, providing the opportunity to reuse the simulation software directly on the agent control level. The platform was implemented in Java on the top of JADE and includes a library of agents representing the basic material handling components like sensors, conveyors, diverters and work cells. BIB001 apply holonic control to continuous processes such as steel manufacturing, and described the implementation of an agent-based control architectures in two manufacturing process systems, namely a linear, tuneable model for the plastic thermoforming process, and a nonlinear, mathematical and rule-based model for the metal powder grinding process. In the machine controller domain, BIB005 apply holonic concepts in the development of machine controllers, which are more flexible and open than traditional NC technology. This holonic behavior is supported by advanced planning, execution and monitoring actions. BIB020 focused on the shop floor re-engineering, using agents to represent the physical components which are aggregated into consortia regulated by contracts, achieving agility in the shop floor life-cycle and BIB022 proposes the actorbased assembly system (ABAS) architecture to develop reconfigurable assembly systems in a easy way. presented PS-Bikes as a case study of a multi-agent control system for manufacturing of custom bikes using JADE framework. Self-organization techniques have been inspiring the development of agent-based control solutions. In this field, Hadeli et al. (2004) proposes a self-organization-based coordination and control approach that uses stimergy for the communication among agents and BIB014 use attraction-repulsion fields to implement a production control system based on selforganization principles.

Agent-based distributed manufacturing control: A state-of-the-art survey <s> Industrial implementations <s> At the beginning of the 1990s, the holonic manufacturing paradigm was proposed (among others) to meet the upcoming challenges in the manufacturing industry. Even though the holonic idea is appealing, its implementation would revolutionize the way manufacturing is done today and thus needs to be thoroughly evaluated before it can be adopted. To this end, an industrial feasibility study for (automotive) engine assembly was conducted within the international holonic manufacturing project. During this feasibility study, the deficits of an existing engine assembly system were analyzed and a new assembly layout and control system was proposed. The new assembly system was then compared to the existing assembly system on the basis of a realistic simulation with plant data. The results of the comparison showed that the holonic approach provides robustness and scalability which is unprecedented in existing assembly systems. <s> BIB001 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> Industrial implementations <s> Many modern manufacturing systems are highly automated and are now requiring decentralised 'smart' architectures to control hardware and manage the flow of materials/knowledge, in order to provide responsiveness. This responsiveness is needed to satisfy an ever increasing consumer need for goods that satisfy their unique requirements and are delivered to market both quickly and economically. A key route to achieve this mass-customisation with distributed control is to apply the holonic enterprise paradigm, and one manufacturing process that exhibits a high potential for responsiveness is packaging. Therefore this paper presents some of the main features of such an enterprise - the Holonic Packing Cell demonstrator being built at Cambridge University's Institute for Manufacturing. It must be emphasised that this cell is constructed from state-of-the-art industrial strength facilities to demonstrate a spectrum of responsive manufacturing ideas - it is not built from Lego bricks. <s> BIB002 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> Industrial implementations <s> In this paper, we focus on validation of Multi-Agent System (MAS) behavior. We describe the simulation architecture and the system design methodology to accomplish the appropriate agent behavior for controlling a real-life automation system. The architecture is explained in the context of an industrial-sized water cooling system. Nevertheless, it is intended to operate in a wide spectrum of control domains. In general, after the design of the control system is accomplished, a set of validation procedures takes place. The current needs are to validate both the control and the agent levels as integrated parts. Hence there is a need to establish a general architecture and methodology for easing the commissioning process of the control solution. <s> BIB003 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> Industrial implementations <s> Giving workpieces an active role in searching for processing steps by available machines in a manufacturing system. <s> BIB004

In spite of all the research described above, only few industrial/ laboratorial applications were developed and reported in the literature. and BIB004 use agent technology to design a flexible and robust production system for large series manufacturing that meet rapidly changing operations, designated by Production 2000+, in a factory plant of DaimlerChrysler, producing cylinder heads for four-cylinder diesel engines (used in the Mercedes Benz C and E class 220 CDI). The approach uses agents to represent machines and workpieces, implementing a dynamic resource allocation, similar to the Contract Net Protocol , with the objective of continuous optimization of the throughput. The agent-based control system allows individual work-pieces to be directed dynamically around the production area, by auction off the processing steps that are due first. The built-in redundancy gives the possibility of diverting product to another machine if a breakdown or unavailability occurs. This system was probably the first full-scale industrial agent-based production system that has been brought into operation and the result was a 20% increase in productivity on average . The resulting prototype system was in day-to-day operation for five years up to the end of the life-cycle of the targeted product. Schneider Electric GmbH in co-operation with DaimlerChrysler AG-Research and Technology had developed and implemented a heterogeneous agent-oriented collaborative control system, called FactoryBroker TM , adequate to control widely distributed and heterogeneous devices in environments that are prone to disruptions and where hard real-time constraints are crucial . Another example of agent-based control systems is the Holomobiles BIB001 ) that introduced a new holonic control approach for an assembly system in the automotive industry, namely for assembling engines, which resulted in more robustness and scalability. Holomobiles implemented a control holon for each docking station, engine buffer, machine station and AGV. It uses, as the P2000+ control system, a Contract Net Protocol to request resources, such as machine stations or AGVs. In spite of sharing common aspects with the P2000+ control system, they present significant differences in terms of system design , namely derived from the nature of the production processes (the production of cylinder heads is different from the assembly of engines) and the requirements expected for each production process (flexibility and robustness for the P2000+ system and robustness and scalability for the Holomobiles system). BIB003 developed an agent-based control system for the chilled-water systems and the heating, ventilation and air-conditioned (HVAC) systems of the US Navy ships. The approach is developed according to the foundation of intelligent physical agents (FIPA) specifications and uses dynamic decisionmaking organizations based on agents to plan, commit and execute control tasks. The intelligent agents are used to represent physical devices, such as valves, T-pipe, cooling plants, water services and heat loads, and to represent specific ship functions, corresponding to the ship, chilled-water, power, material handling, heat and ventilation, and combat sub-systems. The agents reside on the PLC controller or on separate hardware, and can contact any device in the network via a job description language (JDL) message. The JDL represents planning, commitment and execution phases during the task negotiation, being used the Contract Net Protocol to establish dynamic negotiations among the agents. Initially, the automation presented in the real ship was simulated in the laboratory using MATLAB/Simulink and then implemented using physical equipment and controllers. BIB002 described the implementation of a holonic packing cell at the assembly cell of University of Cambridge, using the JACK Intelligent Agents TM platform. In this application, the holonic control system is responsible to assembly Gillete TM packages into customer-tailored gift boxes. The boxes are packed by two Fanuc M6i robots and the test bed integrates a storage system and a conveyor system. The approach considers order holons and resource holons that represent the physical components of the system. The designed agent-based system integrates radio frequency identifier (RFID) technology, by using electronic tags embedded in discrete units, replacing the traditional barcodes. This automatic identification allows to uniquely identify goods, thereby enabling decisions to be made by the order holons representing them.

Agent-based distributed manufacturing control: A state-of-the-art survey <s> Integration of paradigms <s> This paper outlines the perspectives opened by the application of the service-orientation paradigm for realizing high-level communications between next-generation, increasingly intelligent embedded devices - it indicates how this approach can benefit the manufacturing industry - it describes the approach adopted by the SIRENA project, as well as the business advantages it is expected to provide - and outlines early experimental results. <s> BIB001 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> Integration of paradigms <s> Manufacturing process planning is the process of selecting and sequencing manufacturing processes such that they achieve one or more goals and satisfy a set of domain constraints. Manufacturing scheduling is the process of selecting a process plan and assigning manufacturing resources for specific time periods to the set of manufacturing processes in the plan. It is, in fact, an optimization process by which limited manufacturing resources are allocated over time among parallel and sequential activities. Manufacturing process planning and scheduling are usually considered to be two separate and distinct phases. Traditional optimization approaches to these problems do not consider the constraints of both domains simultaneously and result in suboptimal solutions. Without considering real-time machine workloads and shop floor dynamics, process plans may become suboptimal or even invalid at the time of execution. Therefore, there is a need for the integration of manufacturing process-planning and scheduling systems for generating more realistic and effective plans. After describing the complexity of the manufacturing process-planning and scheduling problems, this paper reviews the research literature on manufacturing process planning, scheduling as well as their integration, particularly on agent-based approaches to these difficult problems. Major issues in these research areas are discussed, and research opportunities and challenges are identified <s> BIB002 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> Integration of paradigms <s> One of the significant challenges for current and future manufacturing systems is that of providing rapid reconfigurability in order to evolve and adapt to mass customization. This challenge is aggravated if new types of processes and components are introduced, as existing components are expected to interact with the novel entities but have no previous knowledge on how to collaborate. This statement not only applies to innovative processes and devices, but is also due to the impossibility to incorporate knowledge in a single device about all types of available system components. This paper proposes the use of Semantic Web Services in order to overcome this challenge. The use of ontologies and explicit semantics enable performing logical reasoning to infer sufficient knowledge on the classification of processes that machines offer, and on how to execute and compose those processes to carry out manufacturing orchestration autonomously. A series of motivating utilization scenarios are illustrated, and a research roadmap is presented. <s> BIB003

As previously referred, multi-agent systems alone cannot solve all challenges introduced by the demand of reconfigurable production systems. In this way, agent technology must be integrated with other technologies, such as web-based technologies (including web services and Semantic Web) and grid computing for its wide and successful applications in industry in a near future BIB002 . A current challenge is to combine multi-agent systems with service-oriented architectures (SOA) BIB001 BIB003 . The use of standard and open protocols, namely web technologies, e.g. web services, provides a communication platform between distributed and heterogeneous systems and applications. In such service-oriented control architectures, manufacturing resources (e.g. physical devices, software modules, intelligent units, sub-systems) provide services that encapsulate its internal behavior. The control is achieved by orchestrating the services according to a logic behavior model, providing a high-level interface for the composed process. An unanswered problem in multi-agent systems is related to interoperability in heterogeneous environment. This requires significant efforts from the community of researchers in the area of agent-based industrial systems to develop interoperable knowledge-based systems. Semantics and ontologies seem to be the answer to this challenge. The Semantic Web is based on the idea that the data on the web can be defined and linked in such a way that it can be used for the automatic processing and integration of data by the intelligent agents . The Semantic Web aims to provide a common framework that would support programs to share, reuse and process data on the web, particularly when they have been designed independently. The integration with the IEC 61499 function blocks standard , which is a powerful modeling approach in the distributed industrial process control field, seems a suitable solution to develop agent-based real-time and distributed control application. However, in spite of the significant progress in the application of the IEC 61499 function blocks standard, at the moment, none of the major PLC vendors offer design tools and runtime support for the deployment of IEC 61499 function blocks applications. Aiming to meet the requirements of mobility, modularity and re-configurability, new technologies supporting nomad systems, such as wireless networks, embedded systems and RFID systems, must be able to be integrated within agent-based manufacturing control systems.

Agent-based distributed manufacturing control: A state-of-the-art survey <s> Benchmarking <s> Current research in the area of manufacturing planning and control has moved away from traditional centralized solutions towards distributed architectures that range from hierarchical to heterarchical. Between these two extremes of the control architecture spectrum lies the holonic manufacturing systems paradigm, where partial dynamic hierarchies of agents cooperate to meet global system objectives in the face of disturbances. This paper describes a simulation test bed for the evaluation of a distributed multi-agent control architecture for holonic manufacturing systems that integrates discrete-event simulation software into its design to allow the control architecture to be evaluated with a variety of emulated manufacturing systems. <s> BIB001 </s> Agent-based distributed manufacturing control: A state-of-the-art survey <s> Benchmarking <s> This paper presents the development—by the IMS Network of Excellence (cf. http://www. ims-noe.org)—of a web-based benchmarking service for manufacturing control systems. The paper first discusses the rationale behind this development. Next, the architecture and usage of the benchmarking service is presented and illustrated. Finally, the paper addresses the current status of the benchmarking service. <s> BIB002

At the time, in spite of some research reported in the literature, such as BIB001 , and the ongoing work carried out by the special interest group on benchmarking and performance measures of on-line production scheduling systems (see http://www.ims-noe.org) of the IMS network of excellence (IMS-NoE), a benchmark environment is required to provide realistic test cases for the research community test their developed systems, allowing to evaluate and compare different production control approaches. BIB002 developed a web-based benchmarking service for manufacturing control systems that contributes to evaluate and to compare the merits, in terms of performance, efficiency, agility, etc., of different manufacturing control approaches. However, some questions remain unanswered, namely the selection of proper performance indicators, especially those that allow to evaluate qualitative indicators, definition of evaluation campaign, storage of the best practices and easy connection to this service.

A Survey of Internet of Things (IoT) Authentication Schemes † <s> Introduction <s> This paper addresses the Internet of Things. Main enabling factor of this promising paradigm is the integration of several technologies and communications solutions. Identification and tracking technologies, wired and wireless sensor and actuator networks, enhanced communication protocols (shared with the Next Generation Internet), and distributed intelligence for smart objects are just the most relevant. As one can easily imagine, any serious contribution to the advance of the Internet of Things must necessarily be the result of synergetic activities conducted in different fields of knowledge, such as telecommunications, informatics, electronics and social science. In such a complex scenario, this survey is directed to those who want to approach this complex discipline and contribute to its development. Different visions of this Internet of Things paradigm are reported and enabling technologies reviewed. What emerges is that still major issues shall be faced by the research community. The most relevant among them are addressed in details. <s> BIB001 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Introduction <s> Internet of Things (IoT) is slowly but steadily becoming part of different aspects of our lives, with its applications ranging from smart homes, to wearable devices, to healthcare, etc. This wide spectrum of applications results in shared data containing large amount of users' private information. The security of such information becomes a paramount concern. The IoT security requirements include data confidentiality, data integrity, authentication, access control, privacy, etc. In particular, authentication of IoT devices has a particular importance given the variety of attacks which might result from its breach [1]. This paper provides an analysis of the different authentication schemes proposed in the literature. Through a multi-criteria classification, it compares and analyzes the existing authentication protocols, showing their advantages and disadvantages. <s> BIB002 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Introduction <s> Internet of Things (IoT) is slowly but steadily becoming part of different aspects of our lives, with applications ranging from smart homes, to wearable devices, to healthcare, etc. This wide spectrum of applications results in shared data containing large amount of users' private information. The security of such information becomes a paramount concern. The IoT security requirements include data confidentiality, data integrity, authentication, access control, privacy, etc. In particular, authentication of IoT devices has a particular importance given the variety of attacks which might result from its breach [1]. This paper provides a survey of the different authentication schemes proposed in the literature. Through a multi-criteria classification, it compares and analyzes the existing authentication protocols, showing their advantages and disadvantages. <s> BIB003 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Introduction <s> Securing Internet of Things (IoT) systems is a challenge because of its multiple points of vulnerability. A spate of recent hacks and security breaches has unveiled glaring vulnerabilities in the IoT. Due to the computational and memory requirement constraints associated with anomaly detection algorithms in core networks, commercial in-line (part of the direct line of communication) Anomaly Detection Systems (ADSs) rely on sampling-based anomaly detection approaches to achieve line rates and truly-inline anomaly detection accuracy in real-time. However, packet sampling is inherently a lossy process which might provide an incomplete and biased approximation of the underlying traffic patterns. Moreover, commercial routers uses proprietary software making them closed to be manipulated from the outside. As a result, detecting malicious packets on the given network path is one of the most challenging problems in the field of network security. We argue that the advent of Software Defined Networking (SDN) provides a unique opportunity to effectively detect and mitigate DDoS attacks. Unlike sampling-based approaches for anomaly detection and limitation of proprietary software at routers, we use the SDN infrastructure to relax the sampling-based ADS constraints and collect traffic flow statistics which are maintained at each SDN-enabled switch to achieve high detection accuracy. In order to implement our idea, we discuss how to mitigate DDoS attacks using the features of SDN infrastructure. <s> BIB004 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Introduction <s> Abstract Disruptive technological change can contribute to a more abundant world. However, potentially disruptive technologies often struggle to significantly influence practice. One prominent example is additive manufacturing (AM). Although AM is often regarded as the next great technological revolution in waiting, it has not yet established itself on a large scale in many fields of application. We investigate the reasons behind those challenges by looking at the various fields in which AM is applied and relating them to the specific challenges AM faces, as well as the opportunities it offers in those fields. Our findings rely on a multi-perspective technology foresight process that is based on a discourse analytic approach and that comprises data tomography covering the biggest German-language online magazine on AM and qualitative interview data collected from a range of AM stakeholders. The findings provide an empirically well-founded evaluation and explanation of the link between the challenges and opportunities offered by AM and the extent to which this disruptive technology is leveraged in specific fields. The findings prompt recommendations on how new potentially disruptive technologies can foster abundance in traditional, well established market economies based on the example of the well-developed but traditional market economy of Austria. <s> BIB005

The number of connected devices is growing exponentially, forming the so-called Internet of Things (IoT), a large network of networks connecting smart devices such as sensors and actuators. Such devices are adopted in various domains such as public health, smart grids, smart transportation, waste management, smart homes, smart cities, agriculture, energy management, etc. BIB002 BIB003 BIB001 . The requirements and limitations of the connected "things" raise a number of challenges, including connectivity challenges for billions of devices to communicate with each other, security challenges with the need to protect IoT networks from being attacked (according to a Gartner report, 20% of organizations have experienced at least one IoT attack in the last three years BIB005 ) and at the same time from being exploited to become an attack tool (e.g., Mirai botnet BIB004 ). These challenges are "augmented" with the resource-limited nature of IoT devices which renders traditional communication protocols and security schemes inefficient and even infeasible for IoT. The IoT-related security issues are becoming more alarming given the ubiquitousness of IoT devices and their adoption in critical applications, which aggravate the impact of any security breach to the extent of being life-threatening. A sample scenario can be foreseen from a weakness discovered in 2017 The functions of perception, transport (i.e., network layer) and application layers are the same as in the three-layer architecture. The remaining layers of the architecture are:

A Survey of Internet of Things (IoT) Authentication Schemes † <s> 1. <s> The low-cost, off-the-shelf hardware components in unshielded sensor-network nodes leave them vulnerable to compromise. With little effort, an adversary may capture nodes, analyze and replicate them, and surreptitiously insert these replicas at strategic locations within the network. Such attacks may have severe consequences; they may allow the adversary to corrupt network data or even disconnect significant parts of the network. Previous node replication detection schemes depend primarily on centralized mechanisms with single points of failure, or on neighborhood voting protocols that fail to detect distributed replications. To address these fundamental limitations, we propose two new algorithms based on emergent properties (Gligor (2004)), i.e., properties that arise only through the collective action of multiple nodes. Randomized multicast distributes node location information to randomly-selected witnesses, exploiting the birthday paradox to detect replicated nodes, while line-selected multicast uses the topology of the network to detect replication. Both algorithms provide globally-aware, distributed node-replica detection, and line-selected multicast displays particularly strong performance characteristics. We show that emergent algorithms represent a promising new approach to sensor network security; moreover, our results naturally extend to other classes of networks in which nodes can be captured, replicated and re-inserted by an adversary. <s> BIB001 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> 1. <s> Wireless sensor nodes lack hardware support for tamper- resistance and are often deployed in unattended environments, thus leaving them vulnerable to capture and compromise by an adversary. In a node replication attack, an adversary uses the credentials of a compromised node to surreptitiously introduce replicas of that node into the network. These replicas are then used to launch a variety of attacks that subvert the goal of the sensor application, and the operation of the underlying protocols. We present a novel distributed approach called Localized Multicast for detecting node replication attacks. We evaluate the performance and security of our approach both theoretically and via simulation. Our results show that Localized Multicast is more efficient than previous distributed approaches in terms of communication and memory costs. Further, in our approach, the probability of detecting node replicas is much higher than that achieved in previous distributed protocols. <s> BIB002 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> 1. <s> TCP/IP protocols have long been subject to man-in-the-middle (MITM) attacks, but the advent of SSL/TLS was supposed to mitigate that risk for web transactions by providing endpoint authentication and encryption. The advent of Dug Song's 'webmitm' in late 2000 demonstrated the feasibility of mounting an MITM attack on the protocol, but a properlyconfigured client SSL implementation would warn the user about problems with the server certificate. This paper examines the mechanics of the SSL protocol attack, then focusses on the greater risk of SSL attacks when the client is not properly implemented or configured. One faulty SSL client implementation, Microsoft's Internet Explorer, allows for transparent SSL MITM attacks when the attacker has any CA-signed certificate. An even greater risk is posed by unprotected systems where an attacker can preload his/her own trusted root authority certificates. In public environments such as libraries and computer labs, there is little to prevent such an attack from taking place. Casual observation of such places indicates that an attacker would see them as low-risk, high-opportunity environments. <s> BIB003

Processing layer: Also called the middle-ware layer, it is responsible of providing various types of services, mainly storing, analyzing, and processing data with respect to the computational results. Authentication: The process of confirming and insuring the identity of objects. In IoT context, each object should have the ability to identify and authenticate all other objects in the system (or in a given part of the system with which it interacts). Node Capture: Nodes (base node or gateway) can be easily controlled by the attackers. Catching a node empowers an adversary not only to get tightly of cryptographic keys and protocol states, but also to clone and redistribute malicious nodes in the network, which affects the security of the entire network BIB002 BIB001 . Man-in-the-Middle (MITM): According to McAfee , the most recurrent attacks are Denial of Service (DoS) and Man In the Browser (MITB) attacks. This latter, along with the Secure Socket Layer (SSL) attack, which enables attackers to listen to traffic, intercept it, and spoof both ends of the data, constitute the MITM attack BIB003 . Data Accessibility and Authentication: Each application might have many users . Fake or illegal users could have a great impact on the availability of the whole system. Such great number of users means different permission and access control. Authentication factor In sensor-based applications, where sensors (constrained in terms of memory, processing power, battery, etc.) are the main end-devices, the proposed protocols must be lightweight, making a trade-off between power consumption and security. 2. The robustness of authentication protocols against potential attacks, e.g., sybil, node capture, replay, password guessing, message forgery, brute force, man-in-the-middle, DoS, collision, chosen-plaintext, etc., should be considered and analyzed. In particular, it is important to to consider Distributed Denial of Service attacks (second attack against IoT in 2017 as per ).

A Survey of Internet of Things (IoT) Authentication Schemes † <s> 2. <s> Advances in the areas of embedded systems, computing, and networking are leading to an infrastructure composed of millions of heterogeneous devices. These devices will not simply convey information but process it in transit, connect peer to peer, and form advanced collaborations. This ``Internet of Things'' infrastructure will be strongly integrated with the environment, and its integration with the enterprise systems will not only further blur the line between business IT systems and the real world, but will change the way we design, deploy, and use services. New opportunities can emerge for businesses, which can now closely collaborate with the real world. The work presented here proposes an architecture for an effective integration of the Internet of Things in enterprise services. <s> BIB001 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> 2. <s> This paper addresses the Internet of Things. Main enabling factor of this promising paradigm is the integration of several technologies and communications solutions. Identification and tracking technologies, wired and wireless sensor and actuator networks, enhanced communication protocols (shared with the Next Generation Internet), and distributed intelligence for smart objects are just the most relevant. As one can easily imagine, any serious contribution to the advance of the Internet of Things must necessarily be the result of synergetic activities conducted in different fields of knowledge, such as telecommunications, informatics, electronics and social science. In such a complex scenario, this survey is directed to those who want to approach this complex discipline and contribute to its development. Different visions of this Internet of Things paradigm are reported and enabling technologies reviewed. What emerges is that still major issues shall be faced by the research community. The most relevant among them are addressed in details. <s> BIB002 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> 2. <s> The Internet of Things (IoT) emphasizes on connecting every object around us by leveraging a variety of wireless communication technologies. Heterogeneous data fusion is widely considered to be a promising and urgent challenge in the data processing of the IoT. In this study, we first discuss the development of the concept of the IoT and give a detailed description of the architecture of the IoT. And then we design a middleware platform based on service-oriented architecture (SOA) for integration of multisource heterogeneous information. New research angle regarding flexible heterogeneous information fusion architecture for the IoT is the theme of this paper. Experiments using environmental monitoring sensor data derived from indoor environment are performed for system validation. Through the theoretical analysis and experimental verification, the data processing middleware architecture represents better adaptation to multisensor and multistream application scenarios in the IoT, which improves heterogeneous data utilization value. The data processing middleware based on SOA for the IoT establishes a solid foundation of integration and interaction for diverse networks data among heterogeneous systems in the future, which simplifies the complexity of integration process and improves reusability of components in the system. <s> BIB003 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> 2. <s> The paper presents a survey and analysis on the current status and concerns of Internet of things (IoT) security. The IoT framework aspires to connect anyone with anything at anywhere. IoT typically has a three layers architecture consisting of Perception, Network, and Application layers. A number of security principles should be enforced at each layer to achieve a secure IoT realization. The future of IoT framework can only be ensured if the security issues associated with it are addressed and resolved. Many researchers have attempted to address the security concerns specific to IoT layers and devices by implementing corresponding countermeasures. This paper presents an overview of security principles, technological and security challenges, proposed countermeasures, and the future directions for securing the IoT. <s> BIB004 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> 2. <s> This research addresses blackhole routing attacks —a fundamental security attack on the routing of data in IoT networks. Most IoT devices today, from medical devices to connected vehicles and even smart buildings, come bundled with the capability of communicating wirelessly with one another. Consumers are progressively embracing the concept of connected devices while recent studies indicate that security is not high on the priority list of manufacturers especially in the way these devices route and communicate data amongst themselves thus, leaving the door wide open to attacks and compromises. We propose a trust-based RPL routing protocol, which addresses the blackhole attacks. We show that our proposed system is both secure from blackhole attacks while not imposing undue overheads on network traffic. <s> BIB005 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> 2. <s> Data transportation and routing in Internet of Things (IoT) is a challenging issue where massive data collection and gathering are predictable. The Routing Protocol for Low- power and Lossy Networks (RPL) is one of the best candidates to ensure routing in 6LoWPAN networks. However, RPL is vulnerable to a number of attacks related to exchanged control messages. In this paper, we propose a new secure routing protocol based on RPL referred to as Secure-RPL (SRPL). The main aim of SRPL is to prevent misbehaving nodes from maliciously changing control message values such as the rank of a node that may disturb a network by creating a fake topology. We introduce the concept of rank threshold along with hash chain authentication technique to deal with internal attacks like sinkhole, black hole, selective forwarding attacks etc. Simulation results show that SRPL is robust and resistant to this kind of attacks based on malicious manipulation of RPL metrics. <s> BIB006 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> 2. <s> The Internet of Things (IoT) is defined as a paradigm in which objects equipped with sensors, actuators, and processors communicate with each other to serve a meaningful purpose. In this paper, we survey state-of-the-art methods, protocols, and applications in this new emerging area. This survey paper proposes a novel taxonomy for IoT technologies, highlights some of the most important technologies, and profiles some applications that have the potential to make a striking difference in human life, especially for the differently abled and the elderly. As compared to similar survey papers in the area, this paper is far more comprehensive in its coverage and exhaustively covers most major technologies spanning from sensors to applications. <s> BIB007 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> 2. <s> Internet of Things will connect millions of things to the Internet to make our lives more convenient. However, Internet of Things security is an essential factor. OAuth is one of the most successful authentication and authorization protocols on the Internet. This article proposes push OAuth and personal OAuth authorization server by expanding OAuth for a secure access to the information on Internet of Things devices. In personal OAuth, the smartphones that communicate with remote servers to deliver information on Internet of Things devices can be the OAuth authorization server. Hospitals (OAuth client) that intend to access the information on Internet of Things devices cannot know millions of OAuth authorization server when the smartphone becomes the OAuth authorization server. This article proposes the push OAuth that changes the OAuth protocol and issues the OAuth token when the OAuth authorization server registers to the OAuth client first. Personal OAuth authorization server is far more trustworthy th... <s> BIB008 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> 2. <s> Every day, a new technology comes up and the primary reason why it fails to attract many people in this era is the concern of privacy and security. Each day, along with the new technology comes a load of vulnerabilities waiting to be exploited. IoT (Internet of Things) is the latest trend and like all technology, it is open for exploitation. The most common attack which is used to bring down a whole network, without even finding a loophole in the security — DoS can be used to pull down any IoT network as well. In this paper, we propose a honeypot model for mitigating DoS attacks launched on IoT devices. Honeypots are commonly used in online servers as a decoy to the main server so that the attack is mitigated to the decoy instead of the main server. Here a similar methodology is used to avoid the whole IoT system from being shut down due to a DoS attack. <s> BIB009 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> 2. <s> Abstract Disruptive technological change can contribute to a more abundant world. However, potentially disruptive technologies often struggle to significantly influence practice. One prominent example is additive manufacturing (AM). Although AM is often regarded as the next great technological revolution in waiting, it has not yet established itself on a large scale in many fields of application. We investigate the reasons behind those challenges by looking at the various fields in which AM is applied and relating them to the specific challenges AM faces, as well as the opportunities it offers in those fields. Our findings rely on a multi-perspective technology foresight process that is based on a discourse analytic approach and that comprises data tomography covering the biggest German-language online magazine on AM and qualitative interview data collected from a range of AM stakeholders. The findings provide an empirically well-founded evaluation and explanation of the link between the challenges and opportunities offered by AM and the extent to which this disruptive technology is leveraged in specific fields. The findings prompt recommendations on how new potentially disruptive technologies can foster abundance in traditional, well established market economies based on the example of the well-developed but traditional market economy of Austria. <s> BIB010 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> 2. <s> Internet of Things (IoT) devices is rapidly becoming ubiquitous while IoT services are becoming pervasive. Their success has not gone unnoticed and the number of threats and attacks against IoT devices and services are on the increase as well. Cyber-attacks are not new to IoT, but as IoT will be deeply interwoven in our lives and societies, it is becoming necessary to step up and take cyber defense seriously. Hence, there is a real need to secure IoT, which has consequently resulted in a need to comprehensively understand the threats and attacks on IoT infrastructure. This paper is an attempt to classify threat types, besides, analyze and characterize intruders and attacks facing IoT devices and services. <s> BIB011

Business layer: Its work covers the overall IoT system actions and functionality. The application layer sends the data to the business layer whose role is to build business models, graphs, and flowcharts to analyze data, in order to play a role in decision making about business strategies and road-maps. Other architectures can also be identified in the literature. In BIB003 BIB001 , the authors used a five-layer architecture based on Service Oriented Architecture (SOA) that helps the integration of IoT in enterprise services. In BIB007 , the authors considered a non-layered approach for the architecture (e.g., cloud architecture, fog architecture, social IoT, and architecture based on human brain processing). For the rest of this paper, we consider the three-layer architecture. The authorization: The process of giving permission to an entity to do or have something BIB008 . BIB002 . Integrity: The way toward keeping up the consistency, precision and dependability of information over its whole life cycle. In IoT, the alteration of basic information or even the infusion of invalid information could prompt major issues, e.g., in smart health systems use cases it could lead to the death of the patient BIB011 . BIB010 . Confidentiality: The process of ensuring that the information is only accessed by authorized people. Two main issues should be considered BIB004 regarding confidentiality in IoT: firstly, to ensure that the object receiving the data is not going to move/transfer these data to other objects and, secondly, to consider the data management. Denial of Service (DoS) Attack: A type of attacks that shuts down the system or network and prevents authorized users from accessing it. This could be achieved by overwhelming the system or network with large amount of spam requests all at the same time, thus overloading the system and preventing it from delivering the normal service BIB009 . Denial of Service (DoS): This type of attacks occurs also at the network layer by jamming the transmission of radio signals, using a fake node, affecting the transmission or routing of data between nodes BIB005 BIB006 . Data privacy and identity: The fact that IoT connects different devices from different manufacturers leads to the application of different authentication schemes. The integration of these schemes is a challenging issue to ensure data privacy and identity.

A Survey of Internet of Things (IoT) Authentication Schemes † <s> Security Services <s> Many new "networkable" devices, which constitute the Internet of Things, are low energy and lightweight. These devices must devote most of their available energy and computation to executing core application functionality, making the task of affordably supporting security and privacy quite challenging. <s> BIB001 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Security Services <s> The paper presents a survey and analysis on the current status and concerns of Internet of things (IoT) security. The IoT framework aspires to connect anyone with anything at anywhere. IoT typically has a three layers architecture consisting of Perception, Network, and Application layers. A number of security principles should be enforced at each layer to achieve a secure IoT realization. The future of IoT framework can only be ensured if the security issues associated with it are addressed and resolved. Many researchers have attempted to address the security concerns specific to IoT layers and devices by implementing corresponding countermeasures. This paper presents an overview of security principles, technological and security challenges, proposed countermeasures, and the future directions for securing the IoT. <s> BIB002 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Security Services <s> Today, the world is influenced by new emerging technologies. As a result we are surrounded by a number of smart devices. These smart devices make our life easy and convenient. On the contrary, we are exposed to a number of threats and cyber attacks. There is always a threat to our privacy. In this paper, we discuss the different applications of IOT and the security threats involved. <s> BIB003

As previously mentioned, the use of connecting objects in everyday people's lives can make security issues life-threatening. The smartness integrated into homes, cars, and electric grids can be diverted into harmful scenarios when exploited by hackers. Different hacking scenarios presented in the past years BIB001 illustrate the level of harm that could result from a security breach, especially with the development and large adoption of IoT applications dealing with sensitive information (personal, industrial, governmental, etc.). The main IoT security concerns are: authentication, authorization, integrity, confidentiality, non-repudiation, Availability, and privacy BIB003 BIB002 .

A Survey of Internet of Things (IoT) Authentication Schemes † <s> 5. <s> The emerging Internet-of-Things (IoT) are vulnerable to Sybil attacks where attackers can manipulate fake identities or abuse pseudoidentities to compromise the effectiveness of the IoT and even disseminate spam. In this paper, we survey Sybil attacks and defense schemes in IoT. Specifically, we first define three types Sybil attacks: SA-1, SA-2, and SA-3 according to the Sybil attacker’s capabilities. We then present some Sybil defense schemes, including social graph-based Sybil detection (SGSD), behavior classification-based Sybil detection (BCSD), and mobile Sybil detection with the comprehensive comparisons. Finally, we discuss the challenging research issues and future directions for Sybil defense in IoT. <s> BIB001 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> 5. <s> The Internet of Things (IoT) comprises a diversity of heterogeneous objects that collects data in order to disseminate information to applications. The IoT data dissemination service can be tampered by several types of attackers. Among these, the Sybil attack emerged as the most critical since it operates in the data confidentiality. Although there are approaches against Sybil attack in several services, they disregard the presence of heterogeneous devices and have complex solutions. This paper presents a study highlighting strengths and weaknesses of Sybil attack detection approaches when applied in the IoT content dissemination. An evaluation of the LSD solution was made to assess its effectiveness and efficiency in a IoT network. <s> BIB002

Non-repudiation: The way toward guaranteeing the ability to demonstrate that a task or event has occurred (and by whom), with the goal that this cannot be denied later. In other words, the object cannot deny the authenticity of a specific data transferred. . Availability: The process of ensuring that the service needed is available anywhere and anytime for the intended users. This includes in IoT, the availability of the objects themselves. Fake Node/Sybil Attack: A type of attacks where the attacker can deploy fake identities using fake nodes. With the presence of a sybil node, the whole system might generate wrong data or even the neighbor nodes will receive spam data and will mislay their privacy BIB001 BIB002 . The fake nodes could be used to transmit data to "legitimate" nodes leading them to consume their energy, which could lead the whole service to go down. IoT layer: Indicates the layer at which the authentication procedure is applied. • Perception layer: Responsible for collecting, processing, and digitizing information perceived data by the end nodes in IoT platform.

A Survey of Internet of Things (IoT) Authentication Schemes † <s> 3. <s> The Internet of Things (IoT) will feature pervasive sensing and control capabilities via a massive deployment of machine-type communication (MTC) devices. The limited hardware, low-complexity, and severe energy constraints of MTC devices present unique communication and security challenges. As a result, robust physical-layer security methods that can supplement or even replace lightweight cryptographic protocols are appealing solutions. In this paper, we present an overview of low-complexity physical-layer security schemes that are suitable for the IoT. A local IoT deployment is modeled as a composition of multiple sensor and data subnetworks, with uplink communications from sensors to controllers, and downlink communications from controllers to actuators. The state of the art in physical-layer security for sensor networks is reviewed, followed by an overview of communication network security techniques. We then pinpoint the most energy-efficient and low-complexity security techniques that are best suited for IoT sensing applications. This is followed by a discussion of candidate low-complexity schemes for communication security, such as on–off switching and space-time block codes. The paper concludes by discussing open research issues and avenues for further work, especially the need for a theoretically well-founded and holistic approach for incorporating complexity constraints in physical-layer security designs. <s> BIB001 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> 3. <s> Wake-up-radio-based sensing systems make use of radio- triggering techniques and ultra-low power wake-up receivers (WuRs) to enable on-demand asynchronous network wake ups. Thanks to this, they have the potential to achieve low latency data collection at minimum energy cost, thus meeting the challenging lifetime and quality-of-service demands of emerging Internet of Things (IoT) and Wireless Sensor Networks (WSNs) applications. However, the fact that nodes can be remotely activated on-demand makes wake-up-radio-based networks vulnerable to energy exhausting attacks. In this paper, with a focus on practical implementation and validation, we present a full-fledged solution to counteract Denial-of-Sleep (DoS) attacks to wake-up-radio-based sensing systems. A core component of our proposed solution is a key exchange protocol based on Elliptic Curve Cryptography (the Fully Hashed MQV protocol), which we use in conjunction with implicit certificates. <s> BIB002 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> 3. <s> Many of the benefits of an Internet of Things sensor network model stem from the extremely long service life of its base sensing layer. When data from the base sensing layer is provided by very low power technologies, such as Bluetooth Low Energy, a class of vulnerabilities called Denial of Sleep attacks can be especially devastating to the network. These attacks can reduce the lifespan of the sensing nodes by several orders of magnitude, rendering the network largely unusable. This paper investigates a Denial of Sleep attack against the Bluetooth Low Energy protocol that allows a malicious actor to rapidly drain the battery of a targeted sensing node, including power analysis, simulation results, and an example implementation. The outcome will be utilized to build better defenses and more predictable environments. <s> BIB003

Denial of Sleep Attack: One of the essential objective of an IoT network is the capability of sensing through an extensive number of distributed nodes, each providing small data, such as temperature, humidity, vibration, etc., at a set interval and then going to sleep for another time interval in order to allow the nodes to operate for long service life. The denial of sleep attack works on the power supply of the node with a significant goal to increase the power consumption in order to reduce the service lifetime of the node by preventing the node from going asleep after sending the appropriate sensed data BIB002 BIB003 . Eavesdropping/sniffing: This type of passive attacks gives the intruder the ability to listen to the private communication over the communication link BIB001 . The intruder might be able to extract useful information such as usernames and passwords, node identification or node configuration, which could lead to other types of attacks, e.g., fake node, replay attack, etc. Dealing with the Availability of Big data: IoT connects a huge number of end devices, which leads to a huge amount of data to be managed. This causes an overhead on the application to analyze this data, which has a big impact on the availability of the service(s) provided by the application. Regarding the security requirements for the application layer, authentication is required while protecting the privacy of users (respectively, data). In addition, there should be an information security management scheme that includes resource management and physical security information management. Table 1 gives a summary of the security requirements of the three-layers in the IoT architecture. In Table 1 , it is clear that authentication is a core security mechanism that should applied at different layers. An IoT use case might need an authentication between the end devices and an intermediate device (gateway). The gateway should authenticate itself while sending data to the cloud, and the application (mobile or web) should be authenticated to the cloud in order to collect data for analysis. Authentication procedure • One-way authentication: In a scenario of two parties wishing to communicate with each other, only one party will authenticate itself to the other, while the other one remains unauthenticated. • Two-way authentication: It is also called mutual authentication, in which both entities authenticate each other. • Three-way authentication: Where a central authority authenticates the two parties and helps them to mutually authenticate themselves. There is a need to consider location and identity privacy in certain IoT applications especially smart grids and VANETs. 4. The communication overhead of authentication protocols is a key factor, especially when dealing with power-limited devices; the number of messages exchanged between authentication parties should be kept as low as possible. In the same context, the size of the messages should be as small as possible due to the restricted bandwidth of the wireless communication protocols used. 5. Low computation cost should be considered while designing IoT authentication schemes especially for power-constrained and processing-limited IoT environment. This emphasizes the need to adopt lightweight cryptographic algorithms and protocols while designing authentication solutions. 6. IoT authentication scheme should be scalable in the sense that it should manage large number of nodes s well as have the ability to add new nodes without any further setup or configuration. 7. Authentication service should be ensured for the three layers of IoT architecture (application, network and perception layer). 8. Heterogeneity of devices in IoT networks must be taken into consideration while designing IoT authentication schemes. 9. Hardware security using "PUF" is the current trend due to its advantages over software security approaches. A combination between software solutions (lower cost) and hardware solutions (more secure) should be considered.

A Survey of Internet of Things (IoT) Authentication Schemes † <s> 4. <s> Wireless sensor network (WSN) is an emerging technology that shows great promise for various futuristic applications both for mass public and military. The sensing technology combined with processing power and wireless communication makes it lucrative for being exploited in abundance in future. The inclusion of wireless communication technology also incurs various types of security threats. The intent of this paper is to investigate the security related issues and challenges in wireless sensor networks. We identify the security threats, review proposed security mechanisms for wireless sensor networks. We also discuss the holistic view of security for ensuring layered and robust security in wireless sensor networks. <s> BIB001 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> 4. <s> AbstractIn wormhole attacks, attackers create a low-latency link between two points in the network. This can be achieved by either compromising two or more sensor nodes of the network or adding a new set of mali-cious nodes to the network. Once the link is established, the attacker collects data packets on one end of the link, sends the data packets using the low-latency link and replays them at the other end. Wormhole attacks result in alterations in network data flow thereby deceiving the base station. Although implementing a wormhole attack is relatively simple, detecting it is not a trivial task as the replayed information is usually valid. This paper focuses on wormhole attacks and presents the state-of-the-art in wormhole attack detection in wireless sensor networks. The existing wormhole detection protocols are presented in detail and, based on the existing research, the open research areas and future research directions in wormhole attack detection are provided. <s> BIB002 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> 4. <s> This paper seeks to investigate the performance of the Exponentially Weighted Moving Average (EWMA) for mining big data and detection of DDoS attacks in Internet of Things (IoT) infrastructure. The paper will investigate the trade­off between the algorithm's detection rate, false alarm and detection delay. The paper seeks to further investigate how the performance of the algorithm is affected by the tuning parameters and how various network attack intensity affect its performance. The performance results are analyzed and discussed and further suggestion is also discussed. <s> BIB003 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> 4. <s> There is a concern about possible threats deriving from the widespread adoption of IoT (Internet of Things). The number of devices connected to the Internet is going to increase dramatically, potentiating their security risks. A Distributed Denial of Service (DDoS) attack is a good candidate to explore IoT security vulnerabilities, because of the enormous number of new devices connected to the Internet there is also an increase in the number of possible compromised devices. This study aims to analyze the efficiency of a DDoS attack in a typical IoT environment, by using simulations that, in the best of our knowledge, have not been conducted yet. <s> BIB004 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> 4. <s> “Wireless Sensor Networks (WSNs) are currently using in various application areas like as military, health care, ecological, medical, process management, environmental, industrial, agriculture etc. In this article we discussed about the security issues in WSNs. First of all we constraints on the obstacles in WSNs, requirement for security and different attacks at various levels with their solutions in WSNs. In the WSNs the security issues manage by these categories: cryptography techniques, a key management schemes, secure routing protocols, secure data aggregation and intrusion detection. Along with this we also compare the various routing protocols and security mechanism at high level and low level management in WSNs”. At last we also discussed the open research issues and challenges in WSNs. <s> BIB005 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> 4. <s> Securing Internet of Things (IoT) systems is a challenge because of its multiple points of vulnerability. A spate of recent hacks and security breaches has unveiled glaring vulnerabilities in the IoT. Due to the computational and memory requirement constraints associated with anomaly detection algorithms in core networks, commercial in-line (part of the direct line of communication) Anomaly Detection Systems (ADSs) rely on sampling-based anomaly detection approaches to achieve line rates and truly-inline anomaly detection accuracy in real-time. However, packet sampling is inherently a lossy process which might provide an incomplete and biased approximation of the underlying traffic patterns. Moreover, commercial routers uses proprietary software making them closed to be manipulated from the outside. As a result, detecting malicious packets on the given network path is one of the most challenging problems in the field of network security. We argue that the advent of Software Defined Networking (SDN) provides a unique opportunity to effectively detect and mitigate DDoS attacks. Unlike sampling-based approaches for anomaly detection and limitation of proprietary software at routers, we use the SDN infrastructure to relax the sampling-based ADS constraints and collect traffic flow statistics which are maintained at each SDN-enabled switch to achieve high detection accuracy. In order to implement our idea, we discuss how to mitigate DDoS attacks using the features of SDN infrastructure. <s> BIB006 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> 4. <s> The recent development in wireless communication technology and portable devices has led to the development of Mobile Ad-Hoc Network which has unique features such as dynamic topology, open wireless medium, decentralized system etc but these network are susceptible to various denial of service attacks such as black hole, gray-hole, worm hole etc. In order to mitigate black-hole attack, various protocols have been proposed but these protocols have been not analyzed under gray-hole attack. In this paper, we have made classification of denial of service attack and highlighted the key differences among black hole, sequence number based gray-hole and smart gray-hole attack. We simulated two types of attacks namely Attack1 and Attack2 by modifying AODV protocol. The Attack1 (GAODV) is sequence number based gray-hole attack protocol whereas Attack2 (SGAODV) is smart gray-hole attack protocol. In order to observe the impact of gray-hole attacks on AODV, IDS-AODV and MBDP-AODV, NS-2.35 simulator is used. The simulation results show that MBDP-AODV performs well as compared with IDS-AODV and AODV under sequence number based gray-hole attack. Through simulation, it has also been found that the impact of the smart gray-hole attack is low as compared with the sequence number based gray-hole attack. <s> BIB007

Distributed Denial of Service (DDoS) Attack: A large scale variant of DoS attacks. The most challenging issue is the ability to use the large amount of IoT nodes to pass traffic collected toward the victim server BIB003 BIB004 . There are indications that the DDoS attack called "Mirai" BIB006 occuring on October 2016 benefited from a large number of IoT nodes. Routing attacks: This type of attacks affects how the messages or data are routed. The intruder spoofs, redirects, misdirects or even drops packets at the network layer. The following specific attacks could be considered: (a) Black Hole: It can also be considered as a DoS attack, in which the intruder uses a fake node that welcomes all traffic by asserting that it has the shortest path. As a result, all traffic will be redirected to the fake node that has the ability to redirect them to a proxy server or even drop them . Gray Hole: This type of attacks is similar to the black hole attack but instead of dropping all the packets, it only drops selected ones BIB007 . (c) Worm Hole: In this type of attacks, the intruder creates a connection between two points in the network by either controlling at least two nodes of the network or adding new fake nodes to the network. After forming the link, the intruder collects data from one end and replays them to the other end BIB002 . Hello Flood: The aim of the attacker in this type of attacks is to consume the power of nodes in the system by broadcasting Hello request packets by a fake node to influence all the nodes in the system that they are in the same range, thus causing each one to send packets to its neighbor causing a huge traffic in the network BIB001 . (hello messages are defined in some routing protocols, so that nodes announce themselves to their neighbors.) (e) Sybil: In this attack, a fake node presents multiple identities, thus it can control a considerable part of the framework by being in different places within the network at the same time. When l many sybil nodes are within the same network, they will then send a large amount of information denying the normal nodes from using the network . These potential attacks at the network layer (wired or wireless) lead to the definition of the following security requirements: hop-to-hop encryption, point-to-point authentication, key agreement and management, security routing and intrusion detection BIB005 . Authentication architecture • Distributed: Using a distributed straight authentication method between the communicating parties. • Centralized: Using a centralized server or a trusted third party to distribute and manage the credentials used for authentication. Whether centralized or distributed, the authentication scheme architecture can be: • Hierarchical: Utilizing a multi-level architecture to handle the authentication procedure. No hierarchical architecture is used to deal with the authentication procedure.

A Survey of Internet of Things (IoT) Authentication Schemes † <s> 6. <s> LPWAN (Low Power Wide Area Networks) technologies have been attracting attention continuously in IoT (Internet of Things). LoRaWAN is present on the market as a LPWAN technology and it has features such as low power consumption, low transceiver chip cost and wide coverage area. In the LoRaWAN, end devices must perform a join procedure for participating in the network. Attackers could exploit the join procedure because it has vulnerability in terms of security. Replay attack is a method of exploiting the vulnerability in the join procedure. In this paper, we propose a attack scenario and a countermeasure against replay attack that may occur in the join request transfer process. <s> BIB001 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> 6. <s> Currently one of the most established protocols for machine to machine (M2M) communications is LoRaWAN, designed to provide low power wide area network with features specifically needed to support low-cost, mobile, secure bidirectional communication for the Internet of Things (IoT). In this context security is of pivotal importance, as IoT constitutes a pervasive network of devices highly integrated with our daily life. In this paper we examine key security issues of the procedure used in LoRaWAN to allow an end device to establish a connection with the network server. We have identified vulnerabilities in this protocol, in particular with reference to the use of a random number in the join procedure packet, meant to prevent replay attacks. We first discuss the options that a network server has when detecting a replay attack and then we examine a) the possibility that a legitimate receiver is considered an attacker because of the random number generation issues and b) the possibility for an attacker to exploit this protocol to generate a denial of service (DoS). A wide set of experiments has been conducted using a widely used LoRaWAN chip showing the vulnerabilities of the protocol. <s> BIB002

Replay Attack: In this attack, information is stored and re-transmitted later without having the authority to do that. Such attacks are commonly used against authentication protocols BIB001 BIB002 . Hardware-based: The authentication process might require the use of physical characteristics of the hardware or the hardware itself. • Implicit hardware-based: Uses the physical characteristics of the hardware to enhance the authentication such as Physical Unclonable Function (PUF) or True Random Number Generator (TRNG).

A Survey of Internet of Things (IoT) Authentication Schemes † <s> 8. <s> We propose wireless interceptive Side-Channel Attack (SCA) technique to reveal the 16-byte secret key of the AES-128 encryption algorithm in wireless communications, through Correlation Electromagnetic Analysis (CEMA) for Internet of Things (IoT) applications. The encrypted wireless communication link is established using two ATmega-processor based Arduino boards. There are two key features in our proposed interceptive SCA technique. First, we identify the sensitive modules, which emit significant EM signal (physical leakage information) of the ATmega processor during the encryption process. The significant EM signals are highly correlated with processed data to reveal the secret key. Second, we investigate the resistance of AES-128 encryption algorithm implementation on ATmega processor against CEMA based SCA. The wireless signal is intercepted and correlated with EM signals generated during the encryption process. Based on our experimental results, the correlated EM signals leak out at the three modules — FLASH memory, data bus and SRAM modules during the encryption process are 101.56 dBμV, 105.34 dBμV and 121.79 dBμV respectively. In addition, we perform the CEMA attacks on the AES-128 implementation on the ATmega processor and the secret key is successfully revealed at 20,000 EM traces. <s> BIB001 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> 8. <s> Devices of IoT (Internet of Things) are limited in resources such as CPU, memory etc. The LEA (Lightweight Encryption Algorithm) was standardized as the encryption algorithm suitable for IoT devices in Korea in 2013. However, LEA is vulnerable to the side-channel analysis attack using consumed electric power. To supplement this vulnerability, masking technique is mainly used. However, in case of masking process, the implementation time is increased, losing the characteristics of speedup and lightening. This paper proposes a new and faster LEA algorithm as a countermeasure to the side-channel attack. The proposed algorithm is about 17 times faster than existing algorithms with the masking process to prevent differential side-channel attack. <s> BIB002 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> 8. <s> There are many recent revolutionary advances in information technology that include: wireless communication, the spread of mobile devices, and the Internet-of-Things (IoT). IoT will have an important role in connecting almost everything (mobile devices, cameras, home appliances, healthcare devices, military equipments, …, etc) to the Internet via different communication technologies such as Wi-Fi. This connection will have impact on many sectors of our life such as industry, economy, social life, and ICT sector. Moreover, there will be huge amounts of data (including financial and medical records for example) transmitted between those devices and the non-secure Internet. Some of these data might be very sensitive and their privacy and security must not be compromised. Here comes the need for Cryptographic systems to protect the vital data. There are many hardware and software implementations for the symmetric and asymmetric cryptographic algorithms such as AES, Elliptic Curve Cryptography, and RSA. And since we are talking about protecting physical devices connected to the Internet, we think that the hardware cryptosystems are more useful to be used in this case. In this paper, we introduce the IoT concept, applications, and challenges facing IoT. Then, we present the recent timing and fault Side Channel Attacks on cryptosystem implementations for the most secure encryption algorithms (AES, ECC, and RSA). Also, the countermeasures to protect these cryptosystems from such attacks are also presented. <s> BIB003 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> 8. <s> As networking has become major innovation driver for the Internet of Things as well as Networks on Chips, the need for effective cryptography in hardware is on a steep rise. Both cost and overall system security are the main challenges in many application scenarios, rather than high throughput. In this work we present area-optimized implementations of the lightweight block cipher S imon . All presented cores are protected against side channel attacks using threshold implementation, which applies secret sharing of different orders to prevent exploitable leakages. Implementation results show that, on FPGAs, the higher-order protected S imon core can be smaller than an unprotected AES core at the same security level against classic cryptanalysis. Also, the proposed secure cores consume less than 30 percent the power of any unprotected AES. Security of the proposed cores is validated by provable arguments as well as practical t-test based leakage detection methods. In fact, we show that the first-order protected S imon core does not have first-order leakage and is secure up to 10 million observations against higher-order attacks. The second-order secure implementation could not be exploited at all with up to 100 million observations. <s> BIB004

Side-Channel Attack: This type of attacks occurs on encryption devices by taking advantage of the hardware information where the crypto-system is applied on (chips), such as the execution time, power consumption, power dissipation, and electromagnetic interference produced by electronic devices throughout the encryption procedure. Such information could be analyzed to discover secret keys used during the encryption process BIB001 BIB002 BIB004 BIB003 .

A Survey of Internet of Things (IoT) Authentication Schemes † <s> Application Layer Security Issues and Requirements <s> The paper presents a survey and analysis on the current status and concerns of Internet of things (IoT) security. The IoT framework aspires to connect anyone with anything at anywhere. IoT typically has a three layers architecture consisting of Perception, Network, and Application layers. A number of security principles should be enforced at each layer to achieve a secure IoT realization. The future of IoT framework can only be ensured if the security issues associated with it are addressed and resolved. Many researchers have attempted to address the security concerns specific to IoT layers and devices by implementing corresponding countermeasures. This paper presents an overview of security principles, technological and security challenges, proposed countermeasures, and the future directions for securing the IoT. <s> BIB001 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Application Layer Security Issues and Requirements <s> The rapid development of technology nowadays led people to a new and revolutionary concept, named the Internet of Things. This model imposes that all "objects", such as personal objects (smartphones, notebooks, smart watches, tablets etc), electronic equipment embed with sensors and other environmental elements are always connected to a common network. Therefore, one can access any resource at any time, by using a device recognized in the network. While the IoT may be economically and socially beneficial, the implementation of such a system poses many difficulties, risks and security issues that must be taken into consideration. Nowadays, the architecture of the Internet must be updated and rethought in order to interconnect trillions of devices and to ensure interoperability between them. Nevertheless, the most important problem is the security requirements of the IoT, which is probably one of the main reasons of the relatively slow development of this field. This paper presents the most important application layer protocols that are currently used in the IoT context: CoAP, MQTT, XMPP. We discuss them both separately and by comparison, with focus on the security provided by these protocols. Finally, we provide some future research opportunities and conclusions. <s> BIB002 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Application Layer Security Issues and Requirements <s> In this paper, a comparison of Internet of Things protocols used for data transfer in Internet of Things constrained networks is presented. Setting up such a network with a large number of physical interconnected IoT devices can be a challenge. In the IoT world, one of the key challenges is to efficiently support M2M communication in constrained networks. This can be achieved using MQTT (Message Queuing Telemetry Transport) and CoAP (Constrained Application Protocol) protocols. Choosing the appropriate protocol can be difficult while developing IoT application. There are several conditions that need to be considered while determining which protocol should be used. In this paper, we will evaluate performance and compare these protocols through different scenarios. <s> BIB003

The application layer is responsible for providing services. It hosts a set of protocols for message passing BIB002 BIB003 , e.g., Constrained Application Protocol (COAP), Message Queuing Telemetry Transport (MQTT), Extensible Messaging and Presence Protocol (XMPP), Advanced Message Queuing Protocol (AMQP), etc. This layer directly interacts with the user. Given that the "traditional" application-layer protocols do not perform well within IoT, and since the IoT does not have its own international standards, several security issues arise at the application layer BIB001 .

A Survey of Internet of Things (IoT) Authentication Schemes † <s> Taxonomy of IoT Authentication Schemes <s> Internet of Things (IoT) is slowly but steadily becoming part of different aspects of our lives, with its applications ranging from smart homes, to wearable devices, to healthcare, etc. This wide spectrum of applications results in shared data containing large amount of users' private information. The security of such information becomes a paramount concern. The IoT security requirements include data confidentiality, data integrity, authentication, access control, privacy, etc. In particular, authentication of IoT devices has a particular importance given the variety of attacks which might result from its breach [1]. This paper provides an analysis of the different authentication schemes proposed in the literature. Through a multi-criteria classification, it compares and analyzes the existing authentication protocols, showing their advantages and disadvantages. <s> BIB001 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Taxonomy of IoT Authentication Schemes <s> Internet of Things (IoT) is slowly but steadily becoming part of different aspects of our lives, with applications ranging from smart homes, to wearable devices, to healthcare, etc. This wide spectrum of applications results in shared data containing large amount of users' private information. The security of such information becomes a paramount concern. The IoT security requirements include data confidentiality, data integrity, authentication, access control, privacy, etc. In particular, authentication of IoT devices has a particular importance given the variety of attacks which might result from its breach [1]. This paper provides a survey of the different authentication schemes proposed in the literature. Through a multi-criteria classification, it compares and analyzes the existing authentication protocols, showing their advantages and disadvantages. <s> BIB002

This section presents a taxonomy of IoT authentication schemes using various criteria selected based on the similarities and the main characteristics of these schemes BIB001 BIB002 . As previously mentioned, the authentication can be applied at each of the three layers of the IoT architecture, which makes the diversity of the authentication techniques. These criteria are illustrated in Figure 2 and summarized as follows.

A Survey of Internet of Things (IoT) Authentication Schemes † <s> Use of tokens <s> In this paper, we introduce the first fully implemented two way authentication security scheme for the Internet of Things (IoT) based on existing Internet standards, especially the Datagram Transport Layer Security (DTLS) protocol. The proposed security scheme is based on the most widely used public key cryptography (RSA), and works on top of standard low power communication stacks.We believe that by relying on an established standard, existing implementations, engineering techniques and security infrastructure can be reused, which enables easy security uptake. We present an implemented system architecture for the proposed scheme based on a low-power hardware platform suitable for the IoT. We further demonstrate its feasibility (low overheads and high interoperability) through extensive evaluation. <s> BIB001 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Use of tokens <s> In this paper, we introduce the first fully implemented two-way authentication security scheme for the Internet of Things (IoT) based on existing Internet standards, specifically the Datagram Transport Layer Security (DTLS) protocol. By relying on an established standard, existing implementations, engineering techniques and security infrastructure can be reused, which enables easy security uptake. Our proposed security scheme is therefore based on RSA, the most widely used public key cryptography algorithm. It is designed to work over standard communication stacks that offer UDP/IPv6 networking for Low power Wireless Personal Area Networks (6LoWPANs). Our implementation of DTLS is presented in the context of a system architecture and the scheme's feasibility (low overheads and high interoperability) is further demonstrated through extensive evaluation on a hardware platform suitable for the Internet of Things. <s> BIB002 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Use of tokens <s> Internet of things (IoT) has rapidly become one of the most familiar and perhaps most discussed topic on the research field. The attention for the Internet of Things is mainly due to the new connected products intended to bring greater efficiencies and simplicity to life. Variety of IoT applications lead to equally wide variety of security issues. In this paper, we propose an approach to provide secure authentication mechanism for an IoT network which consists of various kinds of constrained devices using a security manager. This proposed approach protects IoT network from unauthenticated users with security manager using OAuth 2.0 protocol. Moreover, this approach provides flexibility in managing IoT networks. The security manager provides authentication service for multiple IoT networks, which can also help to reduce the cost overhead to maintain secure database in IoT networks. <s> BIB003 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Use of tokens <s> The continuous global growth of Internet of Things devices and deployments constantly produce an equally growing volume of streaming data. This volume puts high pressure on the cloud infrastructure that is expected to manage the storage and processing of such streams. Special purpose IoT data management platforms such as the IoT Framework which is the system under study in this paper, transform raw streaming data into actual information products thus providing value added services to interested users. The purpose of this paper is to evaluate and analyse the impact of the OpenID Connect protocol on the performance of the IoT Framework. The OpenID protocol was introduced in order to enable the IoT framework to be transformed into an IoT Information marketplace. The obtained results show that the introduction of an AA mechanism has a considerable impact on the performance of the IoT Framework. The reason is that the AA server is remote with respect to the IoT platform and therefore the network delay impacts the experienced end to end delay by the user. A possible means to alleviate this problem is to create a local AA server closer to the IoT Platform which either maintains local user AA credentials or federates with other public AA servers. <s> BIB004

• Token-based Authentication: Authenticates a user/device based on an identification token (piece of data) created by a server such as OAuth2 protocol BIB003 or open ID BIB004 . • Non-Token based authentication: Involves the use of the credentials (username/password) every time there is a need to exchange data (e.g., TLS/DTLS BIB001 BIB002 ).

A Survey of Internet of Things (IoT) Authentication Schemes † <s> Smart Grids <s> This paper presents a new password authentication and key-exchange protocol suitable for authenticating users and exchanging keys over an untrusted network. The new protocol resists dictionary attacks mounted by either passive or active network intruders, allowing, in principle, even weak passphrases to be used safely. It also o ers perfect forward secrecy, which protects past sessions and passwords against future compromises. Finally, user passwords are stored in a form that is not plaintext-equivalent to the password itself, so an attacker who captures the password database cannot use it directly to compromise security and gain immediate access to the host. This new protocol combines techniques of zero-knowledge proofs with asymmetric key exchange protocols and o ers signi cantly improved performance over comparably strong extended methods that resist stolen-veri er attacks such as Augmented EKE or B-SPEKE. <s> BIB001 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Smart Grids <s> One of the most important and challenging area in the smart grid context is security and privacy section. Smart grid is a vulnerable system and can be attacked even from aboard, attacks that may cause different level of issues and harms on the devices and society. So, research community has paid attention to this topic and the reasons of required security and privacy for the smart grid. The first step of designing and implementing security for any system such as a smart grid is an authentication scheme followed by a key management protocol. Other security aspects like integrity, authorization and confidentiality can be implemented as long as a strong key management protocol has already been designed and addressed. <s> BIB002 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Smart Grids <s> A smart grid power system is capable of adjusting the amount of electricity generated based on real-time requests from the smart meters of customers, thus avoiding excess electricity generation and facilitating reliable and effective transmission of electricity. To ensure that requests are sent from a valid user, all request messages must be authenticated. On the other hand, by analyzing the electricity usage pattern of a customer, the daily habit of the customer, such as when he is away, may be revealed. Thus, a proper privacy preserving mechanism has to be adopted. This paper attempts to develop a scheme to address these two seemingly contradicting requirements efficiently. By using a tamper-resistant device at the smart appliance and pseudo identities, we derive a privacy preserving authentication scheme to solve the problem. The authentication process is made very efficient by means of Hash-based Message Authentication Code (HMAC). Through simulation, we show that with our scheme, the transmission and signature verification delay induced are very small and the message overhead is only 20 bytes per request message. With our efficient verification process, even under attack, the substation can effectively drop all attack messages, allowing 6 times more valid messages to reach the control center when compared to the case without any verification. Thus our scheme is both efficient and effective. <s> BIB003 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Smart Grids <s> Smart Grid (SG) technology, which aims at bringing the world's aging electric grids into the 21st century by utilizing intelligent transmission and distributed networks, has been gaining momentum in recent years. Despite its attractive features, the SG technology remains vulnerable to some security threats, such as spoofing and man-in-the-middle attacks. In this paper, to address these potential security issues, we propose a light-weight and secure message authentication mechanism. The proposed mechanism is based on Diffie-Hellman key establishment protocol and hash-based message authentication code, which allows various smart meters at different points of the SG to make mutual authentication and achieve message authentication with low latency and few signal message exchanges. Detailed security analysis shows it can satisfy the desirable security requirements. In addition, extensive computer-based simulation also demonstrates its efficiency. <s> BIB004 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Smart Grids <s> Multicast has been envisioned to be useful in many smart grid applications such as demand-response, wide area protection , in-substation protection and various operation and control. Since the multicast messages are related to critical control, authentication is necessary to prevent message forgery attacks. In this paper, we first identify the requirements of multicast communication and multicast authentication in the smart grid. Based on these requirements, we find that one-time signature based multicast authentication is a promising solution, due to its short authentication delay and low computation cost. However, existing one-time signatures are not designed for the smart grid and they may have high storage and bandwidth overhead. To address this problem, we propose a new one-time signature scheme which can reduce the storage cost by a factor of 8 and reduce the signature size by 40% compared with existing schemes. Thus, our scheme is more appropriate for smart grid applications where the receivers have limited storage (e.g., home appliances and field devices) or where data communication is frequent and short (e.g., phasor data). These gains are at the cost of increased computations in signature generation and/or verification and fortunately our scheme can flexibly allocate the computations between the sender and receiver based on their computing resources. We formulate the computation allocation as a nonlinear integer programming problem to minimize the signing cost under a certain verification cost and propose a heuristic solution to solve it. <s> BIB005 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Smart Grids <s> Authentication schemes relying on per-packet signature and per-signature verification introduce heavy cost for computation and communication. Due to its constraint resources, smart grid's authentication requirement cannot be satisfied by this scheme. Most importantly, it is a must to underscore smart grid's demand for high availability. In this paper, we present an efficient and robust approach to authenticate data aggregation in smart grid via deploying signature aggregation, batch verification and signature amortization schemes to less communication overhead, reduce numbers of signing and verification operations, and provide fault tolerance. Corresponding fault diagnosis algorithms are contributed to pinpoint forged or error signatures. Both experimental result and performance evaluation demonstrate our computational and communication gains. <s> BIB006 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Smart Grids <s> Smart grid has emerged as the next generation of power grid, due to its reliability, flexibility, and efficiency. However, smart grid faces some critical security challenges such as the message injection attack and the replay attack. If these challenges cannot be properly addressed, an adversary can maliciously launch the injected or replayed message attacks to degrade the performance of smart grid. To cope with these challenging issues, in this paper, we propose an efficient authentication scheme that employs the Merkle hash tree technique to secure smart gird communication. Specifically, the proposed authentication scheme considers the smart meters with computation-constrained resources and puts the minimum computation overhead on them. Detailed security analysis indicates its security strength, namely, resilience to the replay attack, the message injection attack, the message analysis attack, and the message modification attack. In addition, extensive performance evaluation demonstrates its efficiency in terms of computation complexity and communication overhead. <s> BIB007 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Smart Grids <s> A smart grid (SG) consists of many subsystems and networks, all working together as a system of systems, many of which are vulnerable and can be attacked remotely. Therefore, security has been identified as one of the most challenging topics in SG development, and designing a mutual authentication scheme and a key management protocol is the first important step. This paper proposes an efficient scheme that mutually authenticates a smart meter of a home area network and an authentication server in SG by utilizing an initial password, by decreasing the number of steps in the secure remote password protocol from five to three and the number of exchanged packets from four to three. Furthermore, we propose an efficient key management protocol based on our enhanced identity-based cryptography for secure SG communications using the public key infrastructure. Our proposed mechanisms are capable of preventing various attacks while reducing the management overhead. The improved efficiency for key management is realized by periodically refreshing all public/private key pairs as well as any multicast keys in all the nodes using only one newly generated function broadcasted by the key generator entity. Security and performance analyses are presented to demonstrate these desirable attributes. <s> BIB008 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Smart Grids <s> One-time signatures are known to be one of the most promising solution for time-critical multicast authentication in environments with resource-constrained devices such as smart grid. There have been several solutions for multicast applications and TV-OTS has been known to be the most efficient scheme so far. However, there still exists several challenges for one-time signatures to be deployed in practice. We review existing onetime signatures for multicast authentication and address several challenges such as key management and storage cost problems that can inevitably arise when they are actually deployed in practice. <s> BIB009 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Smart Grids <s> Smart grid network facilitates reliable and efficient power generation and transmission. The power system can adjust the amount of electricity generated based on power usage information submitted by end users. Sender authentication and user privacy preservation are two important security issues on this information flow. In this paper, we propose a scheme such that even the control center (power operator) does not know which user makes the requests of using more power or agreements of using less power until the power is actually used. At the end of each billing period (i.e., after electricity usage), the end user can prove to the power operator that it has really requested to use more power or agreed to use less power earlier. To reduce the total traffic volume in the communications network, our scheme allows gateway smart meters to help aggregate power usage information, and the power generators to determine the total amount of power that needs to be generated at different times. To reduce the impact of attacking traffic, our scheme allows gateway smart meters to help filter messages before they reach the control center. Through analysis and experiments, we show that our scheme is both effective and efficient. <s> BIB010 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Smart Grids <s> Designed an authentication scheme for IoT based smart grid communication.Analyzed the scheme using automated tool ProVerif.The proposed scheme is more lightweight and secure than existing schemes. The Internet of Things (IoT) has plenty of applications including Smart Grid (SG). IoT enables smooth and efficient utilization of SG. It is assumed as the prevalent illustration of IoT at the moment. IP-based communication technologies are used for setting SG communication network, but they are challenged by huge volume of delay sensitive data and control information between consumers and utility providers. It is also challenged by numerous security attacks due to resource constraints in smart meters. Sundry schemes proposed for addressing these problems are inappropriate due to high communication, computation overhead and latency. In this paper, we propose a hybrid Diffie-Hellman based lightweight authentication scheme using AES and RSA for session key generation. To ensure message integrity, the advantages of hash based message authentication code are exploited. The scheme provides mutual authentication, thwarting replay and man-in-the-middle attacks and achieves message integrity, while reducing overall communication and computation overheads. <s> BIB011

Smart grid is taking the momentum over traditional power grids due to its efficiency and effectiveness, but security issues are still challenging in such field. In BIB007 , the authors proposed an authentication scheme based on a Merkle-hash tree. Each home is equipped with a smart meter to collect the consumption of electricity for a time interval and sends the data via wireless communication to the Neighborhood Gateway (NG). The NG sends these data to the control center to collect the bill, which is sent back to the customer. The main contribution is the mutual authentication done between the smart meter and NG using a lightweight scheme that has an efficient computation and communication overhead. The authors of BIB006 also provided an effective and robust approach to authenticate aggregated power usage data in Neighborhood Area Network (NAN) with a fault-tolerance architecture, which is based on digital signature. In BIB002 , the authors proposed a protocol to authenticate Home Area Network (HAN) smart meters with the smart grid utility network, and provided a new approach for key management. While developing a smart grid, two main features should be taken into consideration: the data sent to the control unit or gateway should be sent from a valid smart meter, and there should not be a way of bring out the style of the customer by analyzing his consumption of electricity, thus breaking his privacy. In BIB003 , the authors took the above features into consideration and developed an authentication scheme (called PASS) for smart grids based on Hash-based Message Authentication Code (HMAC). Such approach also ensures the privacy of the customer. To achieve a lightweight message authentication scheme for smart grid, the authors of BIB004 built an approach that allows smart meters to mutually authenticate to other system components and achieve message authentication. The authentication is done using a lightweight Diffie-Hellman and the data integrity is achieved using HMAC. In BIB008 , the authors proposed an authentication protocol for smart grids called Smart Grid Mutual Authentication (SGMA) and another scheme called Smart Grid Key Management (SGKM) for key management. The scheme benefits from the traditional protocols and enhances them to achieve mutual authentication and key management. It uses the Secure Remote Password (SRP) BIB001 that depends on the password entered by the requester to generate a verification ID for further communications, but the enhanced version has less overhead with respect to the exchanged handshake messages and number of packets. Public Key Infrastructure (PKI) is used for key management but due to its overhead regarding the key regeneration, an enhanced version of ID-based Cryptography (EIBC) is used with PKI by replacing the public key with the identity of the requester. In BIB011 , the authors proposed a lightweight authentication protocol for smart grids. It consists of three tiers by using three different protocols for different purposes: Diffie-Hellman is used as key agreement protocol, with the use of RSA and AES for achieving the confidentiality, and HMAC for maintaining message integrity. To address some performance and security challenges such as the storage cost and the key management, the authors of BIB009 discussed one-time signature to be used for multicast authentication in smart grids. The authors of BIB010 proposed a Privacy-preserving Recording and Gateway-assisted Authentication (PRGA) protocol for a hierarchical smart grid network with a gateway-based authentication. The scheme is based on HMAC and homomorphic encryption to authenticate and aggregate the messages sent by smart meters before sending them to the control center. This reduces the amount of exchanged data. The main advantage of such scheme is keeping the identity of the smart meter hidden to the gateway and the control center until the time of generating the bills, thus ensuring privacy. Due to the limitation of one-time signature with respect to the size and the storage of the signature, the authors of BIB005 proposed a one-time signature-based multicast authentication in smart grid. Their scheme deployed a new one-time signature based on a nonlinear integer programming that reduces the computation cost.

A Survey of Internet of Things (IoT) Authentication Schemes † <s> RFID and NFC-Based Applications <s> RFID-tags are becoming very popular tools for identification of products. As they have a small microchip on board, they offer functionality that can be used for security purposes. This chip functionality makes it possible to verify the authenticity of a product and hence to detect and prevent counterfeiting. In order to be successful for these security purposes too, RFID-tags have to be resistant against many attacks, in particular against cloning of the tag. In this paper, we investigate how an RFID-tag can be made unclonable by linking it inseparably to a Physical Unclonable Function (PUF). We present the security protocols that are needed for the detection of the authenticity of a product when it is equipped with such a system. We focus on off-line authentication because it is very attractive from a practical point of view. We show that a PUF based solution for RFID-tags is feasible in the off-line case. <s> BIB001 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> RFID and NFC-Based Applications <s> The Internet of Things (IoT) refers to uniquely identifiable objects (things) which can interact with other objects through the global infrastructure of wireless/wired Internet. The communication technique among a large number of resource-constrained devices that generate large volumes of data has an impact on the security and privacy of the involved objects. In this paper, we propose an encryption method based on XOR manipulation, instead of complex encryption such as using the hash function, for anti-counterfeiting and privacy protection. The enhancement of the security is described and hardware design methodology is also demonstrated. <s> BIB002 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> RFID and NFC-Based Applications <s> The Internet of Things (IoT), an emerging global network of uniquely identifiable embedded computing devices within the existing Internet infrastructure, is transforming how we live and work by increasing the connectedness of people and things on a scale that was once unimaginable. In addition to increased communication efficiency between connected objects, the IoT also brings new security and privacy challenges. Comprehensive measures that enable IoT device authentication and secure access control need to be established. Existing hardware, software, and network protection methods, however, are designed against fraction of real security issues and lack the capability to trace the provenance and history information of IoT devices. To mitigate this shortcoming, we propose an RFID-enabled solution that aims at protecting endpoint devices in IoT supply chain. We take advantage of the connection between RFID tag and control chip in an IoT device to enable data transfer from tag memory to centralized database for authentication once deployed. Finally, we evaluate the security of our proposed scheme against various attacks. <s> BIB003 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> RFID and NFC-Based Applications <s> As one of the core techniques in 5G, the Internet of Things is more interested than ever. Furthermore, radio frequency identification RFID plays a crucial role in Internet of Things development. Although the low-cost RFID system has wide prospect, it has to face with huge challenges because of potential security risks, privacy problems, and efficiency because of its restrictions on processing, storage, and power in RFID tags. One of the possible solutions in secure authentication of the low-cost RFID system is the lightweight RFID authentication protocol. A lightweight RFID mutual authentication protocol with cache in the reader is proposed in this paper, named LRMAPC. The LRMAPC can greatly reduce the computational and transmission cost. Especially, it can reduce computational costs greatly when a large number of tags want to be authenticated. We prove the correctness of LRMAPC using GNY logic. Compared with some existing works, LRMAPC achieves higher efficiency and stronger security. Furthermore, we developed LRMAPC into ULRMAPC, an ultralightweight RFID mutual authentication protocol with cache in the reader. Compared with SASI and Gossamer protocols, ULRMAPC also achieves higher efficiency and stronger security in storage and computation cost. Copyright © 2015 John Wiley & Sons, Ltd. <s> BIB004 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> RFID and NFC-Based Applications <s> As one of the core techniques in 5G, the Internet of Things (IoT) is increasingly attracting people’s attention. Meanwhile, as an important part of IoT, the Near Field Communication (NFC) is widely used on mobile devices and makes it possible to take advantage of NFC system to complete mobile payment and merchandise information reading. But with the development of NFC, its problems are increasingly exposed, especially the security and privacy of authentication. Many NFC authentication protocols have been proposed for that, some of them only improve the function and performance without considering the security and privacy, and most of the protocols are heavyweight. In order to overcome these problems, this paper proposes an ultralightweight mutual authentication protocol, named ULMAP. ULMAP only uses Bit and XOR operations to complete the mutual authentication and prevent the denial of service (DoS) attack. In addition, it uses subkey and subindex number into its key update process to achieve the forward security. The most important thing is that the computation and storage overhead of ULMAP are few. Compared with some traditional schemes, our scheme is lightweight, economical, practical, and easy to protect against synchronization attack. <s> BIB005 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> RFID and NFC-Based Applications <s> With the fast development of the Internet of Things, Radio Frequency Identification (RFID) has been widely applied into many areas. Nevertheless, security problems of the RFID technology are also gradually exposed, when it provides life convenience. In particular, the appearance of a large number of fake and counterfeit goods has caused massive loss for both producers and customers, for which the clone tag is a serious security threat. If attackers acquire the complete information of a tag, they can then obtain the unique identifier of the tag by some technological means. In general, because there is no extra identifier of a tag, it is difficult to distinguish an original tag and its clone one. Once the legal tag data is obtained, attackers can be able to clone this tag. Therefore, this paper shows an efficient RFID mutual verification protocol. This protocol is based on the Physical Unclonable Function (PUF) and the lightweight cryptography to achieve efficient verification of a single tag. The protocol includes three process: tag recognition, mutual verification and update. The tag recognition is that the reader recognizes the tag; mutual verification is that the reader and tag mutually verify the authenticity of each other; update is supposed to maintain the latest secret key for the following verification. Analysis results show that this protocol has a good balance between performance and security. <s> BIB006 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> RFID and NFC-Based Applications <s> Radio frequency identification (RFID) has been considered one of the imperative requirements for implementation of Internet-of-Things applications. It helps to solve the identification issues of the things in a cost-effective manner, but RFID systems often suffer from various security and privacy issues. To solve those issues for RFID systems, many schemes have been recently proposed by using the cryptographic primitive, called physically uncloneable functions (PUFs), which can ensure a tamper-evident feature. However, to the best of our knowledge, none of them has succeeded to address the problem of privacy preservation with the resistance of DoS attacks in a practical way. For instance, existing schemes need to rely on exhaustive search operations to identify a tag, and also suffer from several security and privacy related issues. Furthermore, a tag needs to store some security credentials (e.g., secret shared keys), which may cause several issues such as loss of forward and backward secrecy and large storage costs. Therefore, in this paper, we first propose a lightweight privacy-preserving authentication protocol for the RFID system by considering the ideal PUF environment. Subsequently, we introduce an enhanced protocol which can support the noisy PUF environment. It is argued that both of our protocols can overcome the limitations of existing schemes, and further ensure more security properties. By analyzing the performance, we have shown that the proposed solutions are secure, efficient, practical, and effective for the resource-constraint RFID tag. <s> BIB007

Radio Frequency Identification (RFID) is a wireless technology that consists of tags that can be attached to any physical object or even humans; its main purpose is the identification or detection of the tagged objects. RFID can be deployed in various fields, e.g., supply chain, health care, climate sensing, etc. In BIB006 , the authors suggested a lightweight authentication protocol for RFID tags based on PUF. The protocol consists of three transactions: tag recognition, verification, and update. In the first transaction, the tag reader recognizes the tag. The second transaction is the verification, where the reader and the tag mutually verify the authenticity of each other. In the last transaction (Update), one should keep up the most recent used key for the next verification. To protect the supply chain of connected devices, the authors of BIB003 enabled authentication and perceptibility of the IoT devices, through an RFID-based solution. The authentication process consists of two steps: checking the connectivity between the tag and the IoT device and then approving the perceptibility of the tag. In an IoT-RFID based system, the RFID reader is connected to the Internet to form an IoT end device. On the other side, it is connected to the tagged items via RFID communication protocols. The tagged item is portable and moves from a reader to another, thus there is a need for verifying the identity of each other via authentication. Due to the absence of cryptographic features in RFID, the system is vulnerable to security threats such as impersonation or cloning attacks. In BIB002 , the authors presented an authentication protocol to be used in IoT-RFID use case with the use of lightweight encryption algorithm. To resist against cloning attacks to the RFID tag, the authors of BIB001 proposed an offline authentication for RFID-tags based on PUF. It combined both identification and digital signature security protocols. In the authentication, the tag generates a secret key by challenging the PUF and collecting the response. Such response with the helper data will create a certificate that will be stored inside the ROM of the tag. Next, the verifier authenticates the tag by checking the validity of the certificate. To provide anonymous authentication for RFID systems, the authors of BIB007 presented a PUF-based authentication scheme for classic RFID tags. Then, they provided an enhanced scheme for noisy PUF environment. The main drawback of such scheme is not taking into consideration re-feeding the server with new Challenge-Response Pair (CRP) when the existing pool becomes empty. The authors of proposed a two-factor authentication scheme based on smartphone with Near Field Communication (NFC) feature as first factor and fingerprint of the user as the second factor. Both factors are used to authenticate user on smart library system. The library database then checks if the personal data embedded in the NFC tag and the fingerprint match then the user is authorized to access the internal library network to query for books and providing the user with the rack position (location) in which the book is located. The authors of BIB004 proposed a mutual authentication scheme for IoT RFID-based applications in fifth-generation mobile networks (5G) by providing the reader with a cache, to store the keys (used for authentication) for the recently visited tags, to speed up the authentication, reduce the computation cost and increase the security in storage. The authors of BIB005 proposed a mutual authentication scheme for IoT NFC-based applications in 5G. By using only (lightweight) Shift and XOR operations to suit the performance and storage features of NFC tags, they provided the anonymity of the tags using pseudonym instead of the real identity.

A Survey of Internet of Things (IoT) Authentication Schemes † <s> Vehicular Networks <s> Vehicular communications networks are envisioned for the access to drive-thru Internet and IP-based infotainment applications. These services are supported by roadside access routers (ARs) that connect vehicular ad hoc networks (VANETs) to external IP networks. However, VANETs suffer from asymmetric links due to variable transmission ranges caused by mobility, obstacles, and dissimilar transmission power, which make it difficult to maintain the bidirectional connections and to provide the IP mobility required by most IP applications. Moreover, vehicular mobility results in short-lived connections to the AR, affecting the availability of IP services in VANETs. In this paper, we study the secure and timely handover of IP services in an asymmetric VANET and propose a multihop-authenticated Proxy Mobile IP (MA-PMIP) scheme. MA-PMIP provides an enhanced IP mobility scheme over infrastructure-to-vehicle-to-vehicle (I2V2V) communications that uses location and road traffic information. The MA-PMIP also reacts, depending on the bidirectionality of links, to improve availability of IP services. Moreover, our scheme ensures that the handover signaling is authenticated when V2V paths are employed to reach the infrastructure so that possible attacks are mitigated without affecting the performance of the ongoing sessions. Both analysis and extensive simulations in OMNeT++ are conducted, and the results demonstrate that the MA-PMIP improves service availability and provides secure seamless access to IP applications in asymmetric VANETs. <s> BIB001 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Vehicular Networks <s> Entity authentication and related key management is an active research topic in smart grid security. However, existing works seem to have overlooked the significance that the smart grid is a cyber-physical system, which entails more considerations in the integration of its cyber and physical domains. Ignoring this could possibly undermine security since the effects of cyber authorization in the smart grid are usually extended into the physical domain. The substitution attack, a kind of the man-in-the-middle attack, has been demonstrated using this gap. This paper proposes a two-factor cyber-physical device authentication protocol to defend against coordinated cyber-physical attacks in the smart grid. The idea is to combine a novel contextual factor based on physical connectivity in the power grid with the conventional authentication factor in the challenge-response protocol, widely used in cybersecurity. The resulting protocol provides assurance on not only the digital identity of a device but also the device's controllability in the physical domain. While the design is for the electric vehicle ecosystem, the framework could be readily extended to other smart grid subsystems. <s> BIB002 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Vehicular Networks <s> This paper proposes a new threshold anonymous authentication protocol to support secure and user privacy-preserving communication in vehicular ad hoc networks (VANETs). The proposed protocol holds the following properties: (a) it allows vehicles to broadcast messages with an authenticated while anonymous way; (b) it allows vehicles to accept the received broadcasted messages from others according to their own policies; (c) it allows the trusted third party to reveal the identity information of the vehicle who broadcasted a false message before with constant cost. Due to the above properties, our proposal helps improve the security and performance as compared to the existing authentication protocols for VANETs. <s> BIB003 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Vehicular Networks <s> Vehicular networks are used to coordinate actions among vehicles in traffic by the use of wireless transceivers (pairs of transmitters and receivers). Unfortunately, the wireless communication among vehicles is vulnerable to security threats that may lead to very serious safety hazards. In this work, we propose a viable solution for coping with Man-in-the-Middle attacks. Conventionally, Public Key Infrastructure is utilized for a secure communication with the pre-certified public key. However, a secure vehicle-to-vehicle communication requires additional means of verification in order to avoid impersonation attacks. To the best of our knowledge, this is the first work that proposes to certify both the public key and out-of-band sense-able static attributes to enable mutual authentication of the communicating vehicles. Vehicle owners are bound to preprocess (periodically) a certificate for both a public key and a list of fixed unchangeable attributes of the vehicle. Furthermore, the proposed approach is shown to be adaptable with regards to the existing authentication protocols. We illustrate the security verification of the proposed protocol using a detailed proof in Spi calculus. <s> BIB004 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Vehicular Networks <s> Along with the advancement in wireless communication and vehicle technology, while traveling, exchange of information, has become the most promising research field. Security is central to every major challenge, the mobility model faces nowadays. Proving the identity of one vehicle driver to another during communication, along with message sharing is the main issue to cope with. The main goal of this research is to achieve not only the confidentiality of the message, but also the authentication of vehicle drivers. In this paper, an efficient approach is adopted mainly focuses on the authentication of the vehicle drivers. In this paper vehicle to vehicle protocol is proposed. Steganography BED (Bit-level Extend and Dwindle) technique is used for hiding the message image of variable size behind the cover image. Mapping of Text Message bits with Cover Image decimal values is done, character by character, to generate the new image decimal values and is termed as Even-Odd toggle (E-O Toggle) approach. For authentication and confidentiality purpose, Elliptic Curve Cryptography (ECC) is applied to the user message image, before the embedding with the cover image. <s> BIB005 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Vehicular Networks <s> Existing secure and privacy-preserving schemes for vehicular communications in vehicular ad hoc networks face some challenges, e.g., reducing the dependence on ideal tamper-proof devices, building efficient member revocation mechanisms and avoiding computation and communication bottlenecks. To cope with those challenges, we propose a highly efficient secure and privacy-preserving scheme based on identity-based aggregate signatures. Our scheme enables hierarchical aggregation and batch verification. The individual identity-based signatures generated by different vehicles can be aggregated and verified in a batch. The aggregated signatures can be re-aggregated by a message collector (e.g., traffic management authority). With our hierarchical aggregation technique, we significantly reduce the transmission/storage overhead of the vehicles and other parties. Furthermore, existing batch verification based schemes in vehicular ad hoc networks require vehicles to wait for enough messages to perform a batch verification. In contrast, we assume that vehicles will generate messages (and the corresponding signatures) in certain time spans, so that vehicles only need to wait for a very short period before they can start the batch verification procedure. Simulation shows that a vehicle can verify the received messages with very low latency and fast response. <s> BIB006 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Vehicular Networks <s> In vehicular networks, broadcast communications are critically important, as many safety-related applications rely on single-hop beacon messages broadcast to neighbor vehicles. However, it becomes a challenging problem to design a broadcast authentication scheme for secure vehicle-to-vehicle communications. Especially when a large number of beacons arrive in a short time, vehicles are vulnerable to computation-based Denial of Service (DoS) attacks that excessive signature verification exhausts their computational resources. In this paper, we propose an efficient broadcast authentication scheme called Prediction-Based Authentication (PBA) to not only defend against computation-based DoS attacks, but also resist packet losses caused by high mobility of vehicles. In contrast to most existing authentication schemes, our PBA is an efficient and lightweight scheme since it is primarily built on symmetric cryptography. To further reduce the verification delay for some emergency applications, PBA is designed to exploit the sender vehicle’s ability to predict future beacons in advance. In addition, to prevent memory-based DoS attacks, PBA only stores shortened re-keyed Message Authentication Codes (MACs) of signatures without decreasing security. We analyze the security of our scheme and simulate PBA under varying vehicular network scenarios. The results demonstrate that PBA fast verifies almost 99 percent messages with low storage cost not only in high-density traffic environments but also in lossy wireless environments. <s> BIB007 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Vehicular Networks <s> Most of the existing authentication protocols are based on either asymmetric cryptography like public-key infrastructure (PKI) or symmetric cryptography. The PKI-based authentication protocols are strongly recommended for solving security issues in VANETs. However, they have following shortcomings: (1) lengthy certificates lead to transmission and computation overheads, and (2) lack of privacy-preservation due to revealing of vehicle identity, communicated in broadcasting safety-message. Symmetric cryptography based protocols are faster because of a single secret key and simplicity; however, it does not ensure non-repudiation. In this paper, we present an Efficient, Scalable and Privacy-preserving Authentication (ESPA) protocol for secure vehicular ad hoc networks (VANETs). The protocol employs hybrid cryptography. In ESPA, the asymmetric PKI based pre-authentication and the symmetric hash message authentication code (HMAC) based authentication are adopted during vehicle to infrastructure (V2I) and vehicle to vehicle (V2V) communications, respectively. Extensive simulations are conducted to validate proposed ESPA protocol and compared with the existing work based on PKI and HMAC. The performance analysis showed that ESPA is more efficient, scalable and privacy-preserving secured protocol than the existing work. <s> BIB008 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Vehicular Networks <s> In Vehicular Ad Hoc Networks(VANET), vehicles communicate among each other and with infrastructure points by broadcasting safety and non-safety messages in the network. Due to wireless communication, security and privacy are very important issues to avoid threat in the network. Group based vehicle to vehicle (V2V) communication scheme is proposed here which prevents vehicle from threat. To achieve security and privacy goals, we propose one time authentication for group and then V2V communication is done using group symmetric key within group. Our scheme satisfies all security and privacy requirements such as authentication, non-repudiation and conditional traceability. In case of malicious activity, this scheme can trace malicious vehicle which generates a false message. Computation and communication cost is improved as compared and analyzed with other previous schemes. <s> BIB009 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Vehicular Networks <s> Vehicular ad hoc networks (VANETs) have recently received significant attention in improving traffic safety and efficiency. However, communication trust and user privacy still present practical concerns to the deployment of VANETs, as many existing authentication protocols for VANETs either suffer from the heavy workload of downloading the latest revocation list from a remote authority or cannot allow drivers on the road to decide the trustworthiness of a message when the authentication on messages is anonymous. In this paper, to cope with these challenging concerns, we propose a new authentication protocol for VANETs in a decentralized group model by using a new group signature scheme. With the assistance of the new group signature scheme, the proposed authentication protocol is featured with threshold authentication, efficient revocation, unforgeability, anonymity, and traceability. In addition, the assisting group signature scheme may also be of independent interest, as it is characterized by efficient traceability and message linkability at the same time. Extensive analyses indicate that our proposed threshold anonymous authentication protocol is secure, and the verification of messages among vehicles can be accelerated by using batch message processing techniques. <s> BIB010 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Vehicular Networks <s> Existing secure and privacy-preserving vehicular communication protocols in vehicular ad hoc networks face the challenges of being fast and not depending on ideal tamper-proof devices (TPDs) embedded in vehicles. To address these challenges, we propose a vehicular authentication protocol referred to as distributedaggregate privacy-preserving authentication. The proposed protocol is based on our new multiple trusted authority one-time identity-based aggregate signature technique. With this technique a vehicle can verify many messages simultaneously and their signatures can be compressed into a single one that greatly reduces the storage space needed by a vehicle or a data collector (e.g., the traffic management authority). Instead of ideal TPDs, our protocol only requires realistic TPDs and hence is more practical. <s> BIB011 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Vehicular Networks <s> In vehicular communications, the Vehicular Ad Hoc Network (VANET) is the important role to protect transportation in future. The Broadcast communication is important in vehicular networks, as the safety related applications based on the single hop beacon messages its broadcasts to neighbor vehicles. Since, the dual authentication scheme has critical problem to design a broadcast authentication for secure vehicle-to-vehicle communications. In dual authentication system the vehicles are affects by the computation based Denial of Service (DoS) while the large number of beacons arrive in very short period and this leads to the excessive signature verification which exhaust their computational resources. In this work the Prediction-based Dual Authentication (PBA) is proposed to overcome computation-based DoS attacks, and also resists packet losses caused by high mobility of vehicles. The PBA is the efficient broadcast authentication scheme and also lightweight scheme. However the PBA scheme especially built on symmetric cryptography. In order to reduce the verification delay for some emergency applications, the PBA intend to develop the sender vehicle's can able to predict future beacons in advance. Our proposed scheme only stores the shortened Message Authentication Codes of signature without affecting security to prevent the memory based DoS attacks. To further reduce the verification delay for some emergency time, PBA is intending to develop the sender vehicle's ability to predict future beacons in advance. The dual group key management scheme efficiently broadcast the each valid user in the group and also update frequently if any modification occur in the group. The advantage of our proposed scheme is that the adding / revoking the each user in the VANET group and perform computationally efficient while updating any modification in the group. <s> BIB012 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Vehicular Networks <s> VANETs (Vehicular ad hoc networks) are an important communication to offer safety information and other infotainment applications to vehicles. To secure these networks, it is important to ensure the authentication and confidentiality of communicationV2V (vehicle to vehicle) and V2I (Vehicles to infrastructure). However, in VANET environments, vehicles are moving with a very high speed on the roads, so a strong security can degrade the performance of the protocol. Among the research that presents some schemes to secure VANET communications, we mention the Dual Authentication and Key Management Techniques that suffers from several lack of security. In this paper, we present an enhancement to this solution to obtain a deliberate compromise between quality of service and security. Security analysis proves that our scheme is secure and the result of performance validation is promising. <s> BIB013 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Vehicular Networks <s> Electric vehicles are becoming parts of our daily lives with the increasing investment from auto industry. However, their charging is an issue as this requires frequent charging and longer waiting times compared to traditional gasoline-based vehicles. The charging is typically done at residential or public charging stations. With the increased dominance of electric vehicles, one potential solution is to exploit vehicle-to-vehicle charging (V2V) where an electric vehicle can charge another one through a converter-cable assembly. In such cases, however, there needs to be a protocol between the charge supplier and receiver to authenticate each other and authorize the vehicle to open its charging ports. In this paper, we study this problem of authentication and propose a protocol that will utilize key exchange among the users without relying on certificates. We implemented the proposed protocols under WiFi-direct and Bluetooth and demonstrated that the approach can provide the necessary framework of communication before charging starts without any additional overhead. <s> BIB014

Cars nowadays are connected to the Internet or the Internet of Things to form what is called vehicular networks or Internet of Vehicles. Such connectivity is used to provide different services: providing traffic information for users, sharing riding services, charging electrical cars, etc. The field of Electrical Vehicles (EV) is becoming a trend and vehicle authentication is a challenging topic in EV systems. The authors of BIB002 proposed a two-factor authentication scheme for EV, although it can be deployed in different fields. The scheme combines unique contextual feature. The vehicle is connected to the server via a wireless connectivity and to the charger via a charging cable, so it depends to the physical connectivity to verify the identity. The authors of BIB011 proposed an authentication protocol referred to as distributed aggregate privacy-preserving authentication (DAPPA) that can be used to authenticate a vehicle system with the use of multiple trusted authority one-time identity-based aggregate signature technique. A vehicle can verify many messages at the same time, and their signatures can be aggregated into a one message, which decreases the storage space needed by a vehicle or a data collector. In addition, in BIB006 , the authors proposed a protocol to authenticate communication in secure vehicular ad-hoc networks (VANETs) using identity-based aggregate signatures. In BIB007 , the authors provided a broadcast authentication scheme called Prediction-Based Authentication (PBA) that defends against DoS attacks and resists packet loss. The protocol is based on Merkle hash tree construction for instant verification of urgent messages, and self-generated Message Authentication Code (MAC) storage instead of storing all the receiving signatures. As an enhancement to the scheme in BIB007 , the authors of BIB012 provided a Prediction-Based Authentication (PBA) as a broadcast authentication scheme in VANETs. The proposed scheme utilizes both the Elliptic Curve Digital Signature Algorithm (ECDSA) signatures and Time Efficient Stream Loss Tolerant Authentication (TESLA) to authenticate messages between vehicles which proved to be an efficient, effective, and authentication scheme. In BIB013 , an enhancement to Dual Authentication and Key Management Techniques in VANETs is proposed, by providing an authentication phase the first time the vehicle enters the network and a re-authentication phase when the vehicle moves from one coverage area to another without the need to re-iterating the entire verification process. The authors of BIB014 provided a mutual authentication between vehicles using a Challenge Handshake Authentication Protocol (CHAP) that achieves both authentication and authorization, thus allowing vehicle-to-vehicle (V2V) charging using a converter-cable assembly. The authors of BIB008 proposed an authentication protocol for securing VANETs called (ESPA). The protocol uses asymmetric cryptography (PKI) and the symmetric HMAC for vehicle-to-infrastructure (V2I) and vehicle-to-vehicle (V2V) communications. In BIB009 , the authors proposed a new model for authentication in VANETs based on group authentication. Vehicles are grouped and a session key is generated to be used for V2V communication. The authors of BIB003 BIB010 proposed a threshold authentication protocol to support secure and privacy-preserving communications in VANETs. The protocol is described by the use of a group signature scheme for achieving threshold authentication, efficient cancellation, anonymity, and traceability during the communication of vehicles. In BIB004 , the authors used PKI to certify both the public key and a list of fixed unchangeable attributes of the vehicle. In BIB005 , the authors proposed a scheme to prove the identity of one vehicle driver in to another during communication. Elliptic-curve cryptography (ECC) and steganography techniques are used for the authentication and privacy preservation of users. In most VANETs, there was a problem of connecting it to IP networks using road-side access routers (ARs) due to obstacles and mobility of the vehicle. Hence, the authors of BIB001 studied the ability to provide IP connectivity to the vehicles in VANETs using a multi-hop scheme. The location and road traffic information are stored in a location server used for roaming purposes. The approach provides a mutual authentication between the roaming vehicle and ARs.

A Survey of Internet of Things (IoT) Authentication Schemes † <s> Smart Homes <s> Internet of things (IoT) has rapidly become one of the most familiar and perhaps most discussed topic on the research field. The attention for the Internet of Things is mainly due to the new connected products intended to bring greater efficiencies and simplicity to life. Variety of IoT applications lead to equally wide variety of security issues. In this paper, we propose an approach to provide secure authentication mechanism for an IoT network which consists of various kinds of constrained devices using a security manager. This proposed approach protects IoT network from unauthenticated users with security manager using OAuth 2.0 protocol. Moreover, this approach provides flexibility in managing IoT networks. The security manager provides authentication service for multiple IoT networks, which can also help to reduce the cost overhead to maintain secure database in IoT networks. <s> BIB001 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Smart Homes <s> The Internet of Things (IoT) is an emerging trend that generates many new challenges in security and privacy through the interconnect of ubiquitous devices and services over the Internet. Depending on the device and its strict constraints in the IoT, one often needs to refrain from implementing costly public key cryptography to solve the key distribution problem. In this context, Physical Unclonable Functions (PUF) have been identified as a promising replacement that uses already present physical properties of the device. For encryption and authentication services, we combine this technology with Physical Key Generation (PKG) over wireless communication that leverages physical properties of the communications channel. Note that PKG uses same or similar components as PUFs, rendering the combination of both cheap compared to common public-key cryptography such as RSA or ECC. To the best of our knowledge, this is the first approach to combine both technologies to provide confidentiality and authenticity of devices for a lightweight key distribution mechanism. To demonstrate the validity and usefulness of our approach, we derive a generalized architecture for smart home systems and adopt our approach to this scenario. <s> BIB002 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Smart Homes <s> Machine-to-machine M2M techniques have significant application potential in the emerging internet of things, which may cover many fields from intelligence to ubiquitous environment. However, because of the data exposure when transmitted via cable, wireless mobile devices, and other technologies, its security vulnerability has become a great concern during its further extending development. This problem may even get worse if the user privacy and property are considered. Therefore, the authentication process of communicating entities has attracted wide investigation. Meanwhile, the data confidentiality also becomes an important issue in M2M, especially when the data are transmitted in a public and thereby insecure channel. In this paper, we propose a promising M2M application model that connects a mobile user with the home network using the existing popular Time Division-Synchronous Code Division Multiple Access TD-SCDMA network. Subsequently, a password-based authentication and key establishment protocol is designed to identify the communicating parties and hence establish a secure channel for data transmissions. The final analysis shows the reliability of our proposed protocol. Copyright © 2012 John Wiley & Sons, Ltd. <s> BIB003 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Smart Homes <s> Unit and Ubiquitous IoT (U2IoT) is a typical IoT architecture for achieving comprehensive interactions of ubiquitous things. In U2IoT, security becomes a challenging issue due to objects' complicated interactive phases. The object in different phases of being mapped from the physical space to the cyber space has different security requirements. In this work, a concept of Object Life Cycle (OLC) is first proposed to describe the security requirements for the objects' physical access, cyberentity, and physical extinction phase in U2IoT. Meanwhile, considering the variety of objects and related diversity resources, Physical Unclonable Function (PUF) is introduced to design a lightweight mutual authentication protocol for a smart home scenario. The design correctness and formal security are proved based on the BAN-logic. It indicates that the proposed protocol satisfies the main security requirements for the object during the physical access phase. <s> BIB004 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Smart Homes <s> Internet of Things (IoT) comprises of huge number of connected smart devices to exchange sensed data across different networks. With the rapid growth in security applications development, Physical Unclonable Functions (PUF) has gain interest of researchers. PUF exploits the random factors of IC manufacturing process to generate a variety of mismatches. It helps to generate different unique challenge-response pairs to store in repository and then use for secure associations between the smart devices in IoT. Memory based and Analog electronic based PUFs are mostly used for providing dependable security where MOS transistors based PUF achieve a large variety of mismatches with smaller sizes as compared to others. This paper presents a PUF based Authentication Scheme (PAS) along session keying in order to ensure the secure interaction among smart devices in IoT. We have also proposed the registration and authentication mechanism along with session keying based on challenge-response pair. Next, we present the secure command execution protocol among requesting devices, gateway and smart devices. To validate our work, we have setup a testbed and compared the existing schemes with our proposed PAS in terms of communication overhead during authentication and resilience against device capturing. Results ensure the supremacy of proposed approach over the counterparts. <s> BIB005 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Smart Homes <s> Abstract The Internet of Things (IoT) is a vision that broadens the scope of the Internet by incorporating physical objects to identify themselves to the participating entities. This innovative concept enables a physical object to represent itself in the digital world. There have been a lot of speculations and future forecasts about these physical objects connected with the Internet, however, most of them lack secure features and are vulnerable to a wide range of attacks. Miniature sensor nodes, embedded in these physical objects, limit the support for computationally complex and resource-consuming secured algorithms. In this paper, we propose a lightweight mutual authentication scheme for the real-world physical objects of an IoT environment. It is a payload-based encryption scheme which uses a simple four-way handshake mechanism to verify the identities of the participating objects. The real-world objects communicate with each other using the client–server interaction model. Our proposed scheme uses the lightweight features of Constrained Application Protocol (CoAP) to enable the clients to observe resources residing on the server, in an energy-efficient manner. We use Advanced Encryption Standard (AES), with a key length of 128 bits, to establish a secured session for resource observation. We evaluate our scheme for a real-world scenario using NetDuino Plus 2 boards. Our scheme is computationally efficient, incurs less connection overhead and at the same time, provides a robust defence against various attacks such as, resource exhaustion, Denial-of-Service, replay and physical tampering. <s> BIB006

Smart Homes are automated homes, where users will have the ability to control, monitor and access remotely ( using mobile phone (mobile application) or personal computer (web application)) climate systems (heating and air conditioning), appliances, lighting, TV, audio and video systems, etc. The authors of BIB001 developed a security scheme that can be deployed in smart homes. Such approach has the ability to overcome some security threats such as impersonation and replay attack , but it is still vulnerable to eavesdropping. The approach is based on OAuth 2.0 protocol. In BIB002 , the authors introduced a new authentication scheme to authenticate the end devices deployed in smart homes, which is based on the combination of PUF and Physical Key Generation (PKG). The PUF provides the security of the system by generating a secure key based on the physical parameters of the end device ( the design of PUF depends on common circuit fabrication features that give it the ability to create unique secret key). Machine-to-Machine (M2M) communication is taking a lead in the IoT development, but it also has security challenges. In BIB003 , the authors developed a scheme allowing a smart home network remote user to communicate with end devices. The scheme allows the mutual authentication between the communicating parties, besides establishing a secure connection to establish the confidentiality of data. The authors of BIB004 proposed a mutual authentication protocol for IoT end-devices (smart home use case) based on PUF. They introduced the concept of Object Life cycle (OLC) to describe the roadmap of the end-device from manufacture till the deployment in the IoT system and describing the security challenges during this roadmap. The authors of BIB005 developed a PUF-based authentication scheme for IoT devices to provide mutual authentication between the end device and the gateway by using the CRP data stored inside the gateway. It also provides a way for user (smart phone or wearable device) to authenticate itself with the gateway in order to have the ability to communicate with the end devices using session key generated between them. Timestamp data are used by the user to ensure security against replay attacks. The authors of BIB006 proposed a mutual authentication for IoT systems. The scheme is based on the lightweight features of Constrained Application Protocol (CoAP) as an application layer protocol for the communication between client and the server. The secure communication channel is provided by the advantage of Advanced Encryption Standard (AES) cipher. Both the client and the server challenge each other for mutual authentication by encrypting a payload from the message of size 256 bits, and then exchange payloads for verification. The authentication is done during the request-response interaction without the use of an extra layer (DTLS) which increases the communication and computation cost.

A Survey of Internet of Things (IoT) Authentication Schemes † <s> Wireless Sensor Networks <s> Pairwise key establishment is a fundamental security service in sensor networks; it enables sensor nodes to communicate securely with each other using cryptographic techniques. However, due to the resource constraints on sensors, it is infeasible to use traditional key management techniques such as public key cryptography and key distribution center (KDC). To facilitate the study of novel pairwise key predistribution techniques, this paper presents a general framework for establishing pairwise keys between sensors on the basis of a polynomial-based key predistribution protocol [2]. This paper then presents two efficient instantiations of the general framework: a random subset assignment key predistribution scheme and a grid-based key predistribution scheme. The analysis in this paper indicates that these two schemes have a number of nice properties, including high probability (or guarantee) to establish pairwise keys, tolerance of node captures, and low communication overhead. Finally, this paper presents a technique to reduce the computation at sensors required by these schemes. <s> BIB001 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Wireless Sensor Networks <s> To achieve security in wireless sensor networks, it is important to be able to encrypt and authenticate messages sent between sensor nodes. Before doing so, keys for performing encryption and authentication must be agreed upon by the communicating parties. Due to resource constraints, however, achieving key agreement in wireless sensor networks is nontrivial. Many key agreement schemes used in general networks, such as Diffie-Hellman and other public-key based schemes, are not suitable for wireless sensor networks due to the limited computational abilities of the sensor nodes. Predistribution of secret keys for all pairs of nodes is not viable due to the large amount of memory this requires when the network size is large.In this paper, we provide a framework in which to study the security of key predistribution schemes, propose a new key predistribution scheme which substantially improves the resilience of the network compared to previous schemes, and give an in-depth analysis of our scheme in terms of network resilience and associated overhead. Our scheme exhibits a nice threshold property: when the number of compromised nodes is less than the threshold, the probability that communications between any additional nodes are compromised is close to zero. This desirable property lowers the initial payoff of smaller-scale network breaches to an adversary, and makes it necessary for the adversary to attack a large fraction of the network before it can achieve any significant gain. <s> BIB002 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Wireless Sensor Networks <s> AVISPA is a push-button tool for the automated validation of Internet security-sensitive protocols and applications. It provides a modular and expressive formal language for specifying protocols and their security properties, and integrates different back-ends that implement a variety of state-of-the-art automatic analysis techniques. To the best of our knowledge, no other tool exhibits the same level of scope and robustness while enjoying the same performance and scalability. <s> BIB003 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Wireless Sensor Networks <s> Wireless sensor networks (WSN) are typically deployed in an unattended environment, where the legitimate users can login to the network and access data as and when demanded. Consequently, user authentication is a primary concern in this resource-constrained environment before accessing data from the sensor/gateway nodes. In this letter, we present a two-factor user authentication protocol for WSN, which provides strong authentication, session key establishment, and achieves efficiency. <s> BIB004 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Wireless Sensor Networks <s> This letter proposes an efficient group-based handover authentication scheme for mobile WiMAX networks. When the first Mobile Station (MS) of the handover group moves from the service Base Station (BS) to a target BS, the service BS transmits all the handover group members security context to the target BS. Thus the rest of the MSs in the same handover group can bypass the Extensible Authentication Protocol (EAP) and the security context transfer phases to directly perform the handover authentication, which obviously reduces handover latency. Moreover, the proposed scheme not only meets the essential security requirements in handover authentication semantics (such as mutual authentication and resisting the domino effect) but also achieves privacy preservation. <s> BIB005 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Wireless Sensor Networks <s> A wireless medical sensor network (WMSN) can sense humans’ physiological signs without sacrificing patient comfort and transmit patient vital signs to health professionals’ hand-held devices. The patient physiological data are highly sensitive and WMSNs are extremely vulnerable to many attacks. Therefore, it must be ensured that patients’ medical signs are not exposed to unauthorized users. Consequently, strong user authentication is the main concern for the success and large scale deployment of WMSNs. In this regard, this paper presents an efficient, strong authentication protocol, named E-SAP, for healthcare application using WMSNs. The proposed E-SAP includes: (1) a two-factor (i.e., password and smartcard) professional authentication; (2) mutual authentication between the professional and the medical sensor; (3) symmetric encryption/decryption for providing message confidentiality; (4) establishment of a secure session key at the end of authentication; and (5) professionals can change their password. Further, the proposed protocol requires three message exchanges between the professional, medical sensor node and gateway node, and achieves efficiency (i.e., low computation and communication cost). Through the formal analysis, security analysis and performance analysis, we demonstrate that E-SAP is more secure against many practical attacks, and allows a tradeoff between the security and the performance cost for healthcare application using WMSNs. <s> BIB006 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Wireless Sensor Networks <s> Considering the differences between the Internet and the Internet of Things in security problems, we propose an identity authentication scheme based on elliptic curve algorithm for public and private key pair in order to meet the security requirements of the Internet of Things in heterogeneous networking environments. After being simulated with OPNET, the plan proves to be both safe and effective. <s> BIB007 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Wireless Sensor Networks <s> In the centralized Wireless Sensor Network (WSN) architecture there exists a central entity, which acquires, processes and provides information from sensor nodes. Conversely, in the WSN applications in distributed Internet of Things (IoT) architecture, sensor nodes sense data, process, exchange information and perform collaboratively with other sensor nodes and endusers. In order to maintain the trustworthy connectivity and the accessibility of distributed IoT, it is important to establish secure links for end-to-end communication with proper authentication. The authors propose an implicit certificate-based authentication mechanism for WSNs in distributed IoT applications. The developed two-phase authentication protocol allows the sensor nodes and the end-users to authenticate each other and initiate secure connections. The proposed protocol supports the resource scarcity of the sensor nodes, heterogeneity and scalability of the network. The performance and security analysis justify that the proposed scheme is viable to deploy in resource constrained WSNs. <s> BIB008 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Wireless Sensor Networks <s> Abstract The idea of the Internet of Things (IOT) notion is that everything within the global network is accessible and interconnected. As such Wireless Sensor Networks (WSN) play a vital role in such an environment, since they cover a wide application field. Such interconnection can be seen from the aspect of a remote user who can access a single desired sensor node from the WSN without the necessity of firstly connecting with a gateway node (GWN). This paper focuses on such an environment and proposes a novel user authentication and key agreement scheme for heterogeneous ad hoc wireless sensor networks. The proposed scheme enables a remote user to securely negotiate a session key with a general sensor node, using a lightweight key agreement protocol. The proposed scheme ensures mutual authentication between the user, sensor node, and the gateway node (GWN), although the GWN is never contacted by the user. The proposed scheme has been adapted to the resource-constrained architecture of the WSN, thus it uses only simple hash and XOR computations. Our proposed scheme tackles these risks and the challenges posed by the IOT, by ensuring high security and performance features. <s> BIB009 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Wireless Sensor Networks <s> The research is support by National Natural ScienceFoundation of P. R. China ( Grant ::: No.61170065,61373017,61171053,61103195 and 61203217), Peak of Six Major Talent in Jiangsu Province (Grant No.2010DZXX026), the Natural Science Foundation of Jiangsu Province(Grant No.BK2012436 and BK20130882),Scientific & Technological Support Project of Jiangsu ProvinceGrant No. BE2012183 and BE2012755), Natural Science Key Fund for Colleges and Universities in Jiangsu Province (Grant No. ::: 11KJA520001 and 12KJA520002), Scientific Research & Industry Promotion Project for Higher Education InstitutionsJHB2012-7) . <s> BIB010 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Wireless Sensor Networks <s> Abstract Wireless sensor networks (WSN) will play a fundamental role in the future Internet of Things (IoT), with millions of devices actively exchanging confidential information with one another in a multi-hop manner. Ensuring secure end-to-end communication channels is crucial to the success of innovative IoT applications, as they are essential to limit attacks’ impacts and avoid exposure of information. End-to-end security solutions, such as IPsec or DTLS, do not scale well on WSN devices due to limited resources. In this chapter, the optimized two-way authentication solution for tiny devices (TinyTO) combines end-to-end secured communication with WSN design. TinyTO provides confidentiality and integrity within a fast and secure handshake, works with public-key cryptography, and uses elliptic curve cryptography (ECC) for message encryption and authentication. ECC lowers the resource consumption, and suits devices with 10 kByte RAM and 100 kByte ROM. TinyTO does not need a network-wide shared secret, is application-independent, and supports in-network aggregation. <s> BIB011 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Wireless Sensor Networks <s> Spread of wireless network technology has opened new doors to utilize sensor technology in various areas via Wireless Sensor Networks (WSNs). Many authentication protocols for among the service seeker users, sensing component sensor nodes (SNs) and the service provider base-station or gateway node (GWN) are available to realize services from WSNs efficiently and without any fear of deceit. Recently, Li et al. and He et al. independently proposed mutual authentication and key agreement schemes for WSNs. We find that both the schemes achieve mutual authentication, establish session key and resist many known attacks but still have security weaknesses. We show the applicability of stolen verifier, user impersonation, password guessing and smart card loss attacks on Li et al.'s scheme. Although their scheme employs the feature of dynamic identity, an attacker can reveal and guess the identity of a registered user. We demonstrate the susceptibility of He et al.'s scheme to password guessing attack. In both the schemes, the security of the session key established between user and SNs is imperfect due to lack of forward secrecy and session-specific temporary information leakage attack. In addition both the schemes impose extra computational load on resource scanty sensor-nodes and are not user friendly due to absence of user anonymity and lack of password change facility. To handle these drawbacks, we design a mutual authentication and key agreement scheme for WSN using chaotic maps. To the best of our knowledge, we are the first to propose an authentication scheme for WSN based on chaotic maps. We show the superiority of the proposed scheme over its predecessor schemes by means of detailed security analysis and comparative evaluation. We also formally analyze our scheme using BAN logic. We examine recently proposed Li et al.'s and He et al.'s schemes for WSN.We show security weaknesses in both schemes.We propose an improved scheme for WSN using Chebyshev chaotic maps.Formal security proof using BAN logic and conventional analysis assure the security of our scheme.Comparative evaluation shows the superiority of our scheme over related schemes. <s> BIB012 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Wireless Sensor Networks <s> More security concerns and complicated requirements arise in wireless sensor networks than in wired networks, due to the vulnerability caused by their openness. To address this vulnerability, anonymous authentication is an essential security mechanism for preserving privacy and providing security. Over recent years, various anonymous authentication schemes have been proposed. Most of them reveal both strengths and weaknesses in terms of security and efficiency. Recently, Farash et al. proposed a lightweight anonymous authentication scheme in ubiquitous networks, which remedies the security faults of previous schemes. However, their scheme still suffers from certain weaknesses. In this paper, we prove that Farash et al.'s scheme fails to provide anonymity, authentication, or password replacement. In addition, we propose an enhanced scheme that provides efficiency, as well as anonymity and security. Considering the limited capability of sensor nodes, we utilize only low-cost functions, such as one-way hash functions and bit-wise exclusive-OR operations. The security and lightness of the proposed scheme mean that it can be applied to roaming service in localized domains of wireless sensor networks, to provide anonymous authentication of sensor nodes. <s> BIB013 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Wireless Sensor Networks <s> User authentication in wireless sensor networks (WSN) is a critical security issue due to their unattended and hostile deployment in the field. Since the sensor nodes are equipped with limited computing power, storage, and communication modules, authenticating remote users in such resource-constrained environment is a paramount security concern. Until now, impressive efforts have been made for designing authentication schemes with user anonymity by using only the lightweight cryptographic primitives, such as symmetric key encryption/decryption and hash functions. However, to the best of our knowledge, none has succeeded so far. In this paper, we take an initial step to shed light on the rationale underlying this prominent issue. In order to do that here at first, we demonstrate that the existing solutions for anonymous user authentication in WSN are impractical. Subsequently, we propose a realistic authentication protocol for WSN, which can ensure various imperative security properties like user anonymity, untraceability, forward/backward secrecy, perfect forward secrecy, etc. <s> BIB014 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Wireless Sensor Networks <s> The Internet of Things enables the interconnection of smart physical and virtual objects, managed by highly developed technologies. WSN, is an essential part of this paradigm. The WSN uses smart, autonomous and usually limited capacity devices in order to sense and monitor industrial environments. However, if no authentication mechanism is deployed, this system can be accessible, used and controlled by non-authorized users. In this paper, we propose a robust WSN mutual authentication protocol. A real implementation of the protocol was realized on OCARI, one of the most interesting Wireless Sensor Network technologies. All nodes wanting to access the network should be authenticated at the MAC sub-layer of OCARI. This protocol is especially designed to be implemented on devices with low storage and computing capacities. <s> BIB015 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Wireless Sensor Networks <s> By utilizing Internet of Things (IoT), the collected information from the sensor nodes in wireless sensor networks (WSNs) could be provided to the users who are permitted to get access of sensor nodes. As the information from the sensors are transmitted via public network and sensor nodes have limited battery, shift the focus on security and efficiency in WSNs. User authentication is the security task for limiting the access. It is achieved by equipping authorized users with passwords, tokens or biometrics. However, password and token are easy being stolen and forgotten; even biometrics inherit some limitation. More suitable approach is to combine both password and biometric authenticator to harvest benefits in security. This paper proposes a novel authentication and key agreement scheme for WSNs using biohashing. Biohashing facilitates elimination of false accept rates without increasing occurrence of false rejection rate. Additionally, biohashing has highly clear separation of imposter populations and genuine, and zero equal error rate level. The proposed scheme also supports dynamic node addition and user friendly password change mechanism. Using the BAN-logic, we prove that the proposed scheme provides mutual authentication. In addition, we simulate proposed scheme for the security against man-in-the middle attack and replay attack using the AVISPA tool, and the simulation results show that our scheme is safe. Through the informal security analysis, we show that the proposed scheme is secure against the known attacks for authentication protocols. <s> BIB016 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Wireless Sensor Networks <s> With the rapid development of personal information and wireless communication technology, user authentication schemes have been crucial to ensure that wireless communications are secure. As such, various authentication schemes with multi-factor authentication have been proposed to improve the security of electronic communications. Multi-factor authentication involves the use of passwords, smart cards, and various biometrics to provide users with the utmost privacy and data protection. Cao and Ge analyzed various authentication schemes and found that Younghwa An’s scheme was susceptible to a replay attack where an adversary masquerades as a legal server and a user masquerading attack where user anonymity is not provided, allowing an adversary to execute a password change process by intercepting the user’s ID during login. Cao and Ge improved upon Younghwa An’s scheme, but various security problems remained. This study demonstrates that Cao and Ge’s scheme is susceptible to a biometric recognition error, slow wrong password detection, off-line password attack, user impersonation attack, ID guessing attack, a DoS attack, and that their scheme cannot provide session key agreement. Then, to address all weaknesses identified in Cao and Ge’s scheme, this study proposes a security enhanced multi-factor biometric authentication scheme and provides a security analysis and formal analysis using Burrows-Abadi-Needham logic. Finally, the efficiency analysis reveals that the proposed scheme can protect against several possible types of attacks with only a slightly high computational cost. <s> BIB017 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Wireless Sensor Networks <s> Wireless sensor networks (WSNs) are progressive ad hoc networks that comprise of distributed sensors that are typically and randomly deployed over the target region. The valuable information from the sensor nodes is allowed to access with the help of gateway node by the registered user. To ensure secure communication, a session key is exchanged between the participants over the insecure channel. In this paper, we identified some deficiencies in Jung et al.’s scheme and then devised an enhanced biometric-based anonymous user authentication and the key agreement scheme that is also embedded with symmetric cryptosystem for WSNs. The advantage of using biometric login is to ensure the legal user’s efficient login. We conferred about preserving the security of our proposed scheme, we primarily applied formal verification BAN-logic method to check the exactness of mutual authentication. Furthermore, used automated validation of internet security protocols and applications software that is widely accepted and its results confirm that our scheme is secure against active and passive attacks, including forgery, replay, and man-in-the-middle attack. In addition, an informal analysis proves our scheme can withstand various possible attacks on authentication protocols over the insecure channel. Furthermore, our scheme is more appropriate for WSNs based upon the comparison of computational efficiency and security requirements with recent results. <s> BIB018 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Wireless Sensor Networks <s> Abstract Internet of things is a new pattern in wireless communications of things. The basic theory of IoT is the comprehensive presence of different kinds of things in surrounding environment. The three main dimensions of data security in the Internet are integration, privacy and availability. One of the major tools for maintaining the security of Internet medical things is biometric technology. Biometric systems are used in different fields of the smart healthcare. In this regard, we also need a means for identifying people’s identity based on their physical characteristics; the use of biometric data is absolutely critical for IoT functions. Considering this emerging pattern, everything integrated will connect to a virtual chain of connected things with address in the dynamic world of networks. Given the importance of secure connections of things to the Internet and access to them, biometric features are used as identifiers; because they cannot be borrowed, bought, forgotten and they are very difficult to forge or copy. The smart healthcare and IoT resulted in the emergence of a new standard in the applications of biometric data. In this paper, we present a new standard for applying biometric technology to develop smart healthcare using IoT which includes high capacity to access data in addition to being easy to use. In this paper, we reached a more secure way of accessing IoT based on biometrics and fast identity standard by which we can expect significant advances in smart healthcare systems. <s> BIB019

Wireless Sensor Networks (WSN) is the ability to add the connectivity and sensing features to billions of sensors embedded in various fields (appliances at homes, vehicles, grids, etc.). The authors of BIB015 proposed an authentication protocol at the media access control sub-layer called Optimization of Communication for Ad-hoc Reliable Industrial networks (OCARI), in which they used a one-time shared session key. This technique is suitable for resource-constrained devices. Blom key predistribution scheme BIB002 and the polynomial schema BIB001 are considered as appropriate to be used as key management protocols for some IoT use cases. The authors of BIB016 proposed an improved user authentication and key management protocol for WSNs using Bio-hashing BIB017 . The use of the BAN-logic ( logic of belief and action that ensures one part of communication believes that the key in authentication is good) ensures mutual authentication. The authors of BIB011 BIB008 proposed a mutual authentication which consists of two stages: in the enrollment stage, every node should be identified within the system, and in the authentication stage, a number of handshake messages are exchanged between the end device and the server, which result is a session key to be used for upcoming communication. The authors of BIB005 developed a group authentication technique in wireless networks for achieving both mutual authentication and privacy. An end node will have the ability to move between access points without the need for re-authentication every time. All the roaming group members information are transmitted to the Base Station (BS) the first time a Mobile Station (MS) is authenticated. Then, the MS sends the information to the group manager, which aggregates all the information and sends it back to the BS. The BS sends it to the access service network for verifying the authentication. After analyzing some existing chaotic-maps based protocols, the authors of BIB012 provided a chaotic maps-based mutual authentication in WSN. In BIB013 , the authors proposed a lightweight authentication scheme with anonymous features by providing hop-by-hop authentication and un-traceability. The main contribution is the achievement of user privacy in WSN. In BIB014 , the authors provided an authentication protocol for WSN, which can guarantee different security features such as the privacy of the user, un-traceability, backward secrecy, and strong forward secrecy. It is also resilient to node capture and key compromise impersonation attack. To ensure mutual authentication between the user, the end node, and the gateway node (GWN) in a WSN, the authors of BIB009 developed an authentication scheme based on lightweight Hash and XOR operations that gives the remote user the ability to connect to the end nodes in WSN systems without the need to be connected at the beginning to the gateway. Due to the resource-constrained features of the sensor nodes in WSN systems, there is a need for lightweight encryption protocols. Thus, the authors of BIB010 proposed a secure authentication protocol for WSN based on Elliptic Curve Cryptography (ECC) which is considered lightweight compared to the traditional RSA cipher. Besides, the authors applied Attribute-Based Access Control (ABAC) authorization mechanism for access control, which is considered scalable. In BIB007 , the authors proposed an authentication scheme to be deployed in WSNs. The scheme consists of two main steps: the initialization step in which the user is given a public/private key using a lightweight ECC-based protocol. The second step is the authentication step in which two nodes will use the public/private pair to mutually authenticate each other. The authors of BIB006 provided a protocol called E-SAP for Efficient-Strong Authentication Protocol for wireless health care applications. This protocol consists of a number of features: a two-factor authentication (smart card and password), mutual authentication between sensors, symmetric cryptography to ensure message confidentiality and the ability to change passwords. In BIB004 , the authors provided a two-factor-based user authentication scheme for WSN using both passwords and smart cards. The approach is resistant to stolen-verifier and impersonation attacks, but it does not take into consideration node impersonation attack. The authors of proposed a new anonymous authentication scheme for IoT based on the lightweight ECC. The protocol is composed of two main steps: the registration step and the authentication step. A comparison is held between the ECC and RSA with respect to the time spent for encryption and decryption using different sizes of key. The authors of BIB019 presented a secured way for the authentication of devices in Internet of Medical Things (IMT) using the physical characteristics of people for identification. The authors provided a survey of all the biometric techniques used in the literature especially in the smart health environments with an analysis regarding their compatibility with the IMT, a security analysis is also provided. Finally, the authors provided requirements and open issues for researchers and developers while using biometric properties for authentication. The authors of BIB018 proposed a novel mutual authentication and key management for WSN systems based on biometric and symmetric crypto-system. They compared their scheme with the schemes provided in the literature regarding computation and communication cost and security threats using both BAN-logic and AVISPA BIB003 tools.

A Survey of Internet of Things (IoT) Authentication Schemes † <s> Mobile Network and Applications <s> To support Evolved Packet System (EPS) in the Long Term Evolution (LTE) networks, the 3rd Generation Partnership Project (3GPP) has proposed an authentication and key agreement (AKA) protocol, named EPS-AKA, which has become an emerging standard for fourth-generation (4G) wireless communications. However, due to the requirement of backward compatibility, EPS-AKA inevitably inherits some defects of its predecessor UMTS-AKA protocol that cannot resist several frequent attacks, i.e., redirection attack, man-in-the-middle attack, and DoS attack. Meanwhile, there are additional security issues associated with the EPS-AKA protocol, i.e., the lack of privacy-preservation and key forward/backward secrecy (KFS/KBS). In addition, there are new challenges with the emergence of group-based communication scenarios in authentication. In this paper, we propose a secure and efficient AKA protocol, called SE-AKA, which can fit in with all of the group authentication scenarios in the LTE networks. Specifically, SE-AKA uses Elliptic Curve Diffie-Hellman (ECDH) to realize KFS/KBS, and it also adopts an asymmetric key cryptosystem to protect users' privacy. For group authentication, it simplifies the whole authentication procedure by computing a group temporary key (GTK). Compared with other authentication protocols, SE-AKA cannot only provide strong security including privacy-preservation and KFS/KBS, but also provide a group authentication mechanism which can effectively authenticate group devices. Extensive security analysis and formal verification by using proverif have shown that the proposed SE-AKA is secure against various malicious attacks. In addition, elaborate performance evaluations in terms of communication, computational and storage overhead also demonstrates that SE-AKA is more efficient than those existing protocols. <s> BIB001 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Mobile Network and Applications <s> The roaming service enables mobile subscribers to access the internet service anytime and anywhere, which can fulfill the requirement of ubiquitous access for the emerging paradigm of networking, e.g., the Internet of Things (IoT). In this paper, we propose a conditional privacy-preserving authentication with access linkability (CPAL) for roaming service, to provide universal secure roaming service and multilevel privacy preservation. CPAL provides an anonymous user linking function by utilizing a novel group signature technique, which can not only efficiently hide users’ identities but also enables the authorized entities to link all the access information of the same user without knowing the user’s real identity. Specifically, by using the master linking key possessed by the trust linking server, the authorized foreign network operators or service providers can link the access information from the user to improve its service, while preserving user anonymity, e.g., using individual access information to analyze user preferences without revealing user’s identity. Furthermore, the subscribers can also use this functionality to anonymously query their usage of service. In addition, CPAL has an efficient revocation function, which revokes a group of users at the same time. Through extensive analysis, we demonstrate that CPAL resists various security threats and provides more flexible privacy preservation compared to the existing schemes. Meanwhile, performance evaluations demonstrate its efficiency in terms of communication and computation overhead. <s> BIB002 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Mobile Network and Applications <s> With extensive promising applications, machine to machine (M2M) communications or machine-type communication (MTC) have attached a tremendous interest among mobile network operators and research groups. Supporting multiple MTC devices has been considered as an essential requirement in M2M communications. How to achieve a secure and efficient access authentication for a group of MTC devices during roaming is a challenging issue. In this paper, in order to simultaneously resolve the access security and efficiency in MTC, we propose a secure and efficient group roaming scheme for MTC between 3GPP and WiMAX networks, named SEGR, which is characterized by authenticating all MTC devices in a group simultaneously and speeding up the process of authentication through adopting a novel certificateless aggregate signature technique. Through security analysis, the proposed SEGR can provide robust security, especially overcome the drawback of key escrow in identity-based (ID-based) aggregate signature schemes. In addition, performance evaluations in terms of communication overhead and computation complexity demonstrate that SEGR is more efficient than those traditional schemes. <s> BIB003 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Mobile Network and Applications <s> With the pervasiveness of smartphones and the richness of mobile apps, many people are storing increasingly sensitive data on them, in greater quantities. In order to protect this sensitive information from misuse due to loss, or other accidental reasons, strong smartphone authentication has become imperative and has received considerable attention in recent years. However, when we directly implement traditional authentication schemes in smartphone devices, the balance between security and user-friendliness of authentication becomes challenging, mainly because of the input-in-motion environments. In this paper, without adding extra hardware devices, we present a user-friendly, dual-factor authentication scheme called Duth, for Android smartphone devices. Specifically, the proposed Duth scheme is characterized by utilizing the spatial and time features of the user-writing process as two factors of authentication; a user can be authenticated only if these two features are fulfilled. We implement Duth in Java as a library, which we make publicly available. With extensive discussions on parameter selection, we choose proper parameters and implement Duth on a smartphone with Android 2.3 for experiments, and the experiment results demonstrate that Duth can indeed achieve efficient and effective dual-factor authentication. Copyright © 2014 John Wiley & Sons, Ltd. <s> BIB004 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Mobile Network and Applications <s> Purpose ::: ::: ::: ::: ::: The purpose of this paper is to extract the user behaviour and transform it into a unique signature that can be used as implicit authentication technique. Smart devices are equipped with multiple authentication techniques and still remain prone to attacks because all of these techniques require explicit intervention of the user. Entering a pin code, a password or even having a biometric print can be easily hacked by an adversary. ::: ::: ::: ::: ::: Design/methodology/approach ::: ::: ::: ::: ::: In this paper, the authors introduce a novel authentication model to be used as complementary to the existing authentication models. Particularly, the duration of usage of each application and the occurrence time were examined and modelled into a user signature. During the learning phase, a cubic spline function is used to extract the user signature based on his/her behavioural pattern. ::: ::: ::: ::: ::: Findings ::: ::: ::: ::: ::: Preliminary field experiments show a 70 per cent accuracy rate in determining the rightful owner of the device. ::: ::: ::: ::: ::: Originality/value ::: ::: ::: ::: ::: The main contribution of this paper is a framework that extracts the user behaviour and transforms it into a unique signature that can be used to implicitly authenticate the user. <s> BIB005 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Mobile Network and Applications <s> Abstract Long Term Evaluation Advanced (LTE-A) is the third generation partnership project for cellular network that allows subscribers to roam into networks (i.e., the Internet and wireless connections) using spacial purpose base-stations, such as wireless access points and home node B. In such LTE-A based networks, neither base-stations, nor the Internet and wireless connections are trusted because base-stations are operated by un-trusted subscribers. Attackers may exploit these vulnerabilities to violate the privacy of the LTE-A subscribers. On the other hand, the tradeoff between privacy and authentication is another challenge in such networks. Therefore, in this paper, we propose two anonymous authentication schemes based on one-time pseudonymes and Schnorr Zero Knowledge Protocols. Instead of the international mobile subscriber identity, these schemes enable the user equipment, base-stations and mobility management entity to mutually authenticate each others and update the location of the user equipment without evolving the home subscriber server. The security analysis demonstrate that the proposed schemes thwart security and privacy attacks, such as malicious, international mobile subscriber identity catching, and tracking attacks. Additionally, our proposed schemes preserve the location privacy of user equipment since no entity except the mobility management entity and Gate-Way Mobile Location Center can link between the pseudonymes and the international mobile subscriber identity. Also attackers have no knowledge about international mobile subscriber identity. Hence, the proposed schemes achieve backward/forward secrecy. Furthermore, the performance evaluation shows that the proposed handover schemes impose a small overhead on the mobile nodes and it has smaller computation and communication overheads than those in other schemes. <s> BIB006

To allow remote users to access Internet services any time, anywhere, the authors of BIB002 proposed a new scheme to provide a secure roaming for anonymous users benefiting from the group signature technique. They call it Conditional privacy-preserving authentication with access linkability (CPAL). In BIB006 , the authors proposed two authentication schemes, the first one is based on pseudo-random authentication and the second is based on zero-knowledge authentication for providing authentication and location privacy-preserving scheme for LTE-A. The schemes enable all the entities in LTE-A networks to mutually authenticate each other and update their location without involving the subscriber server. The authors of BIB001 provided also a group-based authentication scheme for LTE networks by developing a group temporary key. It is based on both Elliptic Curve Diffie-Hellman (ECDH) that provides forward and backward secrecy, while using asymmetric key protocol to provide user's privacy. The authors of BIB003 proposed SEGR for the authentication of a group of devices using both 3GPP or WIMAX systems. It is based on certificate-less aggregate signature which was proposed to remove the complication of certificate management in public key cryptography. Due to challenging issues in developing a user-friendly authentication scheme for smart phone environment where touch screen is the most user-friendly input peripherals, the authors of BIB004 provided an authentication process for Android smartphone devices using dual-factor authentication called (Duth). The protocol is made up of a registration step in which the place and time of user entering patterns to the touch screen are stored and then the stored data are used as dual factors for authentication. This approach can improve security without adding any extra hardware. The authors of BIB005 proposed a new authentication scheme for mobile phone users based on behavioural pattern. They started by collecting the behavior of mobile phone user regarding the applications used in a specific time and the duration of usage, and then they change these data to a unique pattern to be used as an authentication between the user and the mobile phone.The proposed scheme will be used as complementary to the existing authentication schemes provided by mobile phones (pin code, fingerprint, gestures, etc.).

A Survey of Internet of Things (IoT) Authentication Schemes † <s> Generic IoT Applications <s> In this work we utilize a physically unclonable function (PUF) to improve resilience of authentication protocols to various types of compromise. As an example application, we consider users who authenticate at an ATM using their bank-issued PUF and a password. We present a scheme that is provably secure and achieves strong security properties. In particular, we ensure that (i) the user is unable to authenticate without her device; (ii) the device cannot be used by someone else to successfully authenticate as the user; (iii) the device cannot be duplicated (e.g., by a bank employee); (iv) an adversary with full access to the bank's personal and authentication records is unable to impersonate the user even if he obtains access to the device before and/or after the setup; (v) the device does not need to store any information. We also give an extension that endows the solution with emergency capabilities: if a user is coerced into opening her secrets and giving the coercer full access to the device, she gives the coercer alternative secrets whose use notifies the bank of the coercion in such a way that the coercer is unable to distinguish between emergency and normal operation of the protocol. <s> BIB001 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Generic IoT Applications <s> The security and privacy of the Internet of Things has an impact on the involved stakeholders. Measures ensuring the architecture's resilience to attacks,data confidentiality,access control and client privacy need to be established.A novel transmission model of IoT is proposed with trusted computing technology.And signcryption schemes from bilinear pairings,which realizes the security requirement of IoT in ONS query and object information transmission.A security architecture and security protocols in the EPC ONS query system have been designed in this model.Root-ONS can authenticate the identities and platform creditability of local ONS servers (L-ONS) by trusted authentication server(TAS),and the TAS give a temporary certificate to validated L-ONS who can apply for enquiry services many times with the certificate in the validated time.A security ONS query service with anonymous authentication provides only to those authorized and trusted L-ONS,which prevents the illegal ONS to enquire information of things.In the transmission process,Remote Information Server of Things(R-TIS) wraps the information of things into multiple encryption layers with the routing node's public key according the order of responded nodes from the end to the start.The encryption data is decrypted the outer layer at each routing node,until the Local Information Server of Things(L-TIS) receives the plain text. Meanwhile,the responded nodes can check the integrity of received data and the creditability of routing path in the transmitting procedure.The analysis shows that the novel transmission model of IoT has many properties,such as security,anonymity,trustworthy and attack-resistant. <s> BIB002 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Generic IoT Applications <s> The emergence of Internet of Things (IoT) brings tremendous benefits and opportunities for individuals and businesses. However, there exist several challenges that need to be addressed before a full realization of IoT can be achieved. Fundamentally, IoT enables constant transfer and sharing of data between several “things” (i.e. humans and objects) in order to achieve particular objectives. In such sharing environments, security and privacy of data and messages become important. Authentication, authorization. access control, non-repudiation are important to ensure secure communication in an IoT environment. The lack of computing resources (such as processing power, storage, etc.) and ad-hoc nature of such networks requires researcher to re-think existing techniques to adopt to such environments. In this paper, we propose a framework for authentication, authorization and access control for an IoT environment using capability tokens, PKI and encryption which aims to use minimal computing resources. <s> BIB003 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Generic IoT Applications <s> In this paper, we introduce the first fully implemented two way authentication security scheme for the Internet of Things (IoT) based on existing Internet standards, especially the Datagram Transport Layer Security (DTLS) protocol. The proposed security scheme is based on the most widely used public key cryptography (RSA), and works on top of standard low power communication stacks.We believe that by relying on an established standard, existing implementations, engineering techniques and security infrastructure can be reused, which enables easy security uptake. We present an implemented system architecture for the proposed scheme based on a low-power hardware platform suitable for the IoT. We further demonstrate its feasibility (low overheads and high interoperability) through extensive evaluation. <s> BIB004 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Generic IoT Applications <s> Authentication and key exchange are fundamental techniques for enabling secure communication over mobile networks. In order to reduce implementation complexity and achieve computation efficiency, design issues for efficient and secure biometrics-based remote user authentication scheme have been extensively investigated by research community in these years. Recently, two biometrics-based authentication schemes using smart cards are introduced by Li and Hwang and Li et al., respectively. Li and Hwang proposed an efficient biometrics-based remote user authentication scheme using smart card and Li et al. proposed an improvement. The authors of both schemes claimed that their protocols deliver the important security features and system functionalities, such as without synchronized clock, freely changes password, mutual authentication, as well as low computation costs. However, these two schemes still have much space for security enhancement. In this paper, we first demonstrate a series of vulnerabilities on these two schemes. Then, an enhanced scheme with corresponding remedies is proposed to eliminate all identified security flaws in both schemes. <s> BIB005 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Generic IoT Applications <s> This paper proposed a security framework for the internet of things based on Public Key Infrastructure (PKI) which aims to answer these questions that how to solve the security problem in the internet of things between the client operation and service operation. It illustrates the model by presenting a modified security framework for a variety of applications than traditional PKI framework. Our aim is to show the new security framework in the internet of things that the traditional internet network does not have. <s> BIB006 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Generic IoT Applications <s> In this paper, we introduce the first fully implemented two-way authentication security scheme for the Internet of Things (IoT) based on existing Internet standards, specifically the Datagram Transport Layer Security (DTLS) protocol. By relying on an established standard, existing implementations, engineering techniques and security infrastructure can be reused, which enables easy security uptake. Our proposed security scheme is therefore based on RSA, the most widely used public key cryptography algorithm. It is designed to work over standard communication stacks that offer UDP/IPv6 networking for Low power Wireless Personal Area Networks (6LoWPANs). Our implementation of DTLS is presented in the context of a system architecture and the scheme's feasibility (low overheads and high interoperability) is further demonstrated through extensive evaluation on a hardware platform suitable for the Internet of Things. <s> BIB007 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Generic IoT Applications <s> Cloud computing is an emerging technology that is still unclear to many security problems and user authentication, access control, and ensuring the security of stored data in cloud servers are the most challenging issues in cloud-based environment. Accordingly, this paper offers an efficient and scalable user authentication scheme for cloud computing environment. It the suggested model, various tools and techniques have been introduced and used by using the concept of agent. Therefore, a client-based user authentication agent has been introduced to confirm identity of the user in client-side. Furthermore, a cloud-based software-as-a-service application has been used to confirm the process of authentication for unregistered devices. Moreover, there are two separate servers for storing authentication and cryptography resources from main servers to decrease the dependency of user authentication and encryption processes from main server. Cryptography agent was also introduced to encrypt resources before storing on cloud servers. In overall, the theoretical analysis of the suggested scheme shows that, designing this user authentication and access control model will enhance the reliability and rate of trust in cloud computing environments as an emerging and powerful technology in various industries. <s> BIB008 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Generic IoT Applications <s> Wirelessly connected smart embedded devices, forming the so called Internet of Things network, have achieved unprecedented levels of diffusion as they are adopted in many application domains, ranging from goods transportation to eHealth monitoring infrastructure. As they are always inherently connected, hence exposed to attacks, and as they densely populate our daily life collecting, managing and elaborating data, security has drawn a lot of attention in the literature. In a crowded network, classical security approaches may be not adequate, since they require secret sharing or public key distribution infrastructures. Physically Unclonable Functions (PUFs), introduced so far, are exploitable as security primitives, providing easy authentication and secure key storage mechanisms. Traditional PUF authentication schemes rely on the enrollment of some challenge/response pairs (CRPs), extracted before each device is issued, as it is not feasible to retrieve the whole CRPs set. Moreover, accomplishing such a procedure may introduce a significant overhead due to the cardinality of extracted CRPs and due to size of the device population. To avoid these issues, in this paper we exploit the knowledge of a PUF model in order to make available the whole CRPs set, and, by adopting an encryption scheme, we hide it to avoid model based attacks which can be performed on CRPs sent in clear. To this aim, we show an implementation based on the Anderson PUF and on AES, realized on a Xilinx Zynq-7000 Field Programmable Gate Array. <s> BIB009 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Generic IoT Applications <s> Day by day users has been adopting remote storage system due to its secure service. But some time it may possible that people comes to know what security policy has been applied in server side. As the result, new security risks and attacks are coming in cloud. So, it is necessary to provide more secured and updated authentication scheme. For any cloud application, in which personal or private information are exchanged, single-tier authentication is not sufficient for authentication. In above situation, multi-tier authentication scheme is much more secured than single-tier authentication scheme. In computing environment, there are various multitier authentication schemes, but they do not provide security against insider attacks and virtualization attacks. In cloud environment, whole authentication control lies in the server side. So, it is hard to trust the third party server in cloud system. This paper proposes a secured and more advanced multi-tier authentication scheme for accessing cloud services. <s> BIB010 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Generic IoT Applications <s> Abstract Supporting a massive number of machine to machine (M2M) devices has been considered as an essential requirement in M2M communications. Meanwhile, cyber security is of paramount importance in M2M; if M2M devices cannot securely access the networks through efficient authentication, all applications involving M2M cannot be widely accepted. One of research challenges in M2M is group authentication since a large number of M2M devices accessing the network simultaneously will cause a severe authentication signaling congestion. To solve this problem, as well as reduce authentication overhead of the previous schemes based on public key cryptosystems, we propose a novel lightweight group authentication scheme for resource-constrained M2M (GLARM) under the 3GPP network architecture, which consists of two protocols that can achieve efficient and secure group authentication in the 3GPP access case and non-3GPP access case, respectively. GLARM can not only authenticate all M2M devices simultaneously, but also minimize the authentication overhead. The security analysis shows that the proposed scheme can achieve the security goals, and prevent the various security threats. In addition, performance evaluation demonstrates its efficiency in terms of computation complexity and communication overhead. <s> BIB011 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Generic IoT Applications <s> The article has presented an overview on the promise of PUFs in providing much needed security in IoTs. We have discussed a security violation in the authentication of a commercial-lighting-based IoT, and argued that the use of PUFs as a hardware security primitive can provide a low-cost but secured mechanisThe article has presented an overview on the promise of PUFs in providing much needed security in IoTs. We have discussed a security violation in the authentication of a commercial-lighting-based IoT, and argued that the use of PUFs as a hardware security primitive can provide a low-cost but secured mechanism to subvert such attacks. However, there are several challenges which need to be considered: starting from making the PUF primitives secured against machine learning but yet lightweight, to develop protocols using them to prevent such machine-learning-based attacks. Case studies for a lightweight construction of PUFs (called LSPUF), attacks on it, and a protocol on PUFs, called SPUF, have been detailed to highlight these challenging constraints of PUF-based designs. Finally, a proof-of-concept testbed has been presented using commercially available tools to show how PUFs can be used to interact with other IoT nodes to provide overall security.m to subvert such attacks. However, there are several challenges which need to be considered: starting from making the PUF primitives secured against machine learning but yet lightweight, to develop protocols using them to prevent such machine-learning-based attacks. Case studies for a lightweight construction of PUFs (called LSPUF), attacks on it, and a protocol on PUFs, called SPUF, have been detailed to highlight these challenging constraints of PUF-based designs. Finally, a proof-of-concept testbed has been presented using commercially available tools to show how PUFs can be used to interact with other IoT nodes to provide overall security. <s> BIB012 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Generic IoT Applications <s> Devices in the internet of things (IoT) are frequently (i) resource-constrained, and (ii) deployed in unmonitored, physically unsecured environments. Securing these devices requires tractable cryptographic protocols, as well as cost effective tamper resistance solutions. We propose and evaluate cryptographic protocols that leverage physical unclonable functions (PUFs): circuits whose input to output mapping depends on the unique characteristics of the physical hardware on which it is executed. PUF-based protocols have the benefit of minimizing private key exposure, as well as providing cost-effective tamper resistance. We present and experimentally evaluate an elliptic curve based variant of a theoretical PUF-based authentication protocol proposed previously in the literature. Our work improves over an existing proof-of-concept implementation, which relied on the discrete logarithm problem as proposed in the original work. In contrast, our construction uses elliptic curve cryptography, which substantially reduces the computational and storage burden on the device. We describe PUF-based algorithms for device enrollment, authentication, decryption, and digital signature generation. The performance of each construction is experimentally evaluated on a resource-constrained device to demonstrate tractability in the IoT domain. We demonstrate that our implementation achieves practical performance results, while also providing realistic security. Our work demonstrates that PUF-based protocols may be practically and securely deployed on low-cost resource-constrained IoT devices. <s> BIB013 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Generic IoT Applications <s> The Internet of Things (IoT) represents a great opportunity to connect people, information, and things, which will in turn cause a paradigm shift in the way we work, interact, and think. IoT devices are usually small, low cost, and have limited resources, which makes them vulnerable to physical, side-channel, and cloning attacks. Therefore, any protocol designed for IoT systems should not only be secure but also efficient in terms of usage of chip area, energy, storage, and processing. To address this issue, we present light-weight mutual authentication protocols for IoT systems based on physical unclonable functions. Protocols for two scenarios are presented, one when an IoT device and server wish to communicate and the other when two IoT devices want to establish a session. A security and performance analysis of the protocols shows that they are not only robust against different types of attacks, but are also very efficient in terms of computation, memory, energy, and communication overhead. The proposed protocols are suitable for real time applications and are an attractive choice for implementing mutual authentication in IoT systems. <s> BIB014 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Generic IoT Applications <s> Device authentication is a critical and challenging issue for the emerging Internet of Things (IoT). One promising solution to authenticate IoT devices is to extract a fingerprint to perform device authentication by exploiting variations in the transmitted signal caused by hardware and manufacturing inconsistencies. In this paper, we propose a lightweight device authentication protocol [named speaker-to-microphone (S2M)] by leveraging the frequency response of a speaker and a microphone from two wireless IoT devices as the acoustic hardware fingerprint. S2M authenticates the legitimate user by matching the fingerprint extracted in the learning process and the verification process, respectively. To validate and evaluate the performance of S2M, we design and implement it in both mobile phones and PCs and the extensive experimental results show that S2M achieves both low false negative rate and low false positive rate in various scenarios under different attacks. <s> BIB015 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Generic IoT Applications <s> Internet of Things (IoT) device authentication is weighed as a very important step from security perspective. Privacy and security of the IoT devices and applications is the major issue. From security perspective, important issue that needs to be addressed is the authentication mechanism, it has to be secure from different types of attacks and is easy to implement. The paper gives general idea about how different authentication mechanisms work, and then secure and efficient multi-factor device authentication scheme idea is proposed. The proposed scheme idea uses digital signatures and device capability to authenticate a device. In the proposed scheme device will only be allowed into the network if it is successfully authenticated through multi-factor authentication otherwise the authentication process fails and whole authentication process will restart. By analyzing the proposed scheme idea, it can be seen that the scheme is efficient and has less over head. The scheme not only authenticates the device very efficiently through multi-factor authentication but also authenticates the authentication server with the help of digital signatures. The proposed scheme also mitigates the common attacks like replay and man in the middle because of nonce and timestamp. <s> BIB016 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Generic IoT Applications <s> Abstract There is no doubt that Internet of Things (IoT) occupy a very important role in our daily lives. Indeed, numerous objects that we use every time, are being equipped with electronic devices and protocol suites in order to make them interconnected and connected to the Internet. In IoT, things process and exchange data without human intervention. Therefore, because of this full autonomy, these entities need to recognize and authenticate each other as well as to ensure the integrity of their exchanged data. Otherwise, they will be the target of malicious users and malicious use. Due to the size and other features of IoT, it is almost impossible to create an efficient centralized authentication system. To remedy this limit, in this paper, we propose an original decentralized system called bubbles of trust , which ensures a robust identification and authentication of devices. Furthermore, it protects the data integrity and availability. To achieve such a goal, our approach relies on the security advantages provided by blockchains, and serves to create secure virtual zones ( bubbles ) where things can identify and trust each other. We also provided a real implementation of our mechanism using the C++ language and Ethereum blockchain. The obtained results prove its ability to satisfy IoT security requirements, its efficiency, and its low cost. <s> BIB017 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Generic IoT Applications <s> Internet of Things (IoT) is the interconnection of physical entities to be combined with embedded devices like sensors, activators connected to the Internet which can be used to communicate from human to things for the betterment of the life. Information exchanged among the entities or objects, intruders can attack and change the sensitive data. The authentication is the essential requirement for security giving them access to the system or the devices in IoT for the transmission of the messages. IoT security can be achieved by giving access to authorized and blocking the unauthorized people from the internet. When using traditional methods, it is not guaranteed to say the interaction is secure while communicating. Digital certificates are used for the identification and integrity of devices. Public key infrastructure uses certificates for making the communication between the IoT devices to secure the data. Though there are mechanisms for the authentication of the devices or the humans, it is more reliable by making the authentication mechanism from X.509 digital certificates that have a significant impact on IoT security. By using X.509 digital certificates, this authentication mechanism can enhance the security of the IoT. The digital certificates have the ability to perform hashing, encryption and then signed digital certificate can be obtained that assures the security of the IoT devices. When IoT devices are integrated with X.509 authentication mechanism, intruders or attackers will not be able to access the system, that ensures the security of the devices. <s> BIB018 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Generic IoT Applications <s> Abstract Recently, authentication technologies integrated with the Internet of Things (IoT) and cloud computing have been promptly investigated for secure data retrieval and robust access control on large-scale IoT networks. However, it does not have a best practice for simultaneously deploying IoT and cloud computing with robust security. In this study, we present a novel authentication scheme for IoT-based architectures combined with cloud servers. To pursue the best efficiency, lightweight crypto-modules, such as one-way hash function and exclusive-or operation, are adopted in our authentication scheme. It not only removes the computation burden but also makes our proposed scheme suitable for resource-limited objects, such as sensors or IoT devices. Through the formal verification delivered by Proverif, the security robustness of the proposed authentication scheme is guaranteed. Furthermore, the performance evaluation presents the practicability of our proposed scheme in which a user-acceptable computation cost is achieved. <s> BIB019

The authors of BIB014 provided a protocol for performing authentication between a user and a server, and not between end devices. It is a two-step verification to IoT devices. A password or a shared secret key is considered as the first authentication factor and the use of PUF as the second authentication factor. Due to the large number of devices wishing to access the network which leads to an overload for the authenticating server and in order to achieve mutual authentication and secure key management for resource-constrained devices, the authors in BIB011 provided a group-based lightweight authentication and key agreement scheme called GLARM. GLARM is composed of two fundamental stages: an identification stage and, a group authentication and key-agreement stage based on a combination of message authentication code of a group of devices to be authenticated. In BIB015 , the authors proposed a lightweight device authentication protocol for IoT systems named speaker-to-microphone (S2M). This scheme achieves distance authentication between wireless IoT devices. Is is implemented in both mobile phones and PC to check its experimental results. The author of BIB012 provided a new hardware-based approach, using a hardware fingerprint to authenticate IoT device with its Physical Unclonable Functions (PUF). The author presented the machine learning based attacks on PUF leading to the creation of a software model of the PUF. The author of BIB013 presented an implementation of PUF-based algorithms for IoT device. They used PUF-based elliptic curve for device enrollment, authentication, decryption, and digital signature. The author fingered the environment variations under which the IoT devices will be operating and also the effects of aging in the usage of PUF by using error correction codes. They handled the machine-learning attacks by combining the PUF with the use of ECC to encrypt the generated key. In BIB009 , the authors proposed an authentication mechanism which applies a modeling for the PUF to avoid storing the whole Challenge-Response Pairs (CRPs) or the retrieval of some CRPs during authentication or verification phases, and the model will be hidden by applying a symmetric encryption algorithm (AES) to avoid modeling-based attacks. The authors of BIB010 realized that a single-tier authentication is not adequate for services such as cloud computing. Thus, they proposed a multi-tier authentication scheme without the use of any extra software or hardware. In the first stage, user enters a simple username and password that is validated by the server on the cloud. In the second stage, user pursues a predetermined sequence on the virtual screen during the registration phase. In BIB008 , the authors provided an authentication scheme to be used in cloud computing use cases. They divided the devices into two categories: registered and unregistered devices, and handled the authentication using two different approaches. The registered devices are authenticated using an authentication server. Firstly, the device is registered in the server, then a session key is generated by the server and sent encrypted using Advanced Encryption Standard (AES), to the device to be used for further communications. On the other hand, a cloud-based Software-as-a-Service (SaaS) using modified Diffie-Hellman (DH) algorithm is used to authenticate unregistered devices for accessing cloud services. Using password or smart cards as the only way of authentication for remote users is vulnerable to security attacks. Thus, the authors of BIB005 proposed a novel scheme by using a three-factor authentication. The fingerprint or iris-scan, the smart card, and a password to authenticate a user to remotely-based applications. Although traditional Public Key Infrastructure is considered heavy with respect to the computation and communication cost, the authors of BIB003 BIB006 presented a lightweight PKI to be implemented in IoT use cases. The authors of [145] presented an efficient transmission to ensure privacy of data sent. The proposed scheme used a customized data encapsulation to reduce both the computation and communication overhead. The authors of BIB007 presented a two-way IoT authentication protocol. The authentication is handled using the Datagram Transport Layer Security (DTLS) protocol with the exchange of certificates based on RSA. This work is an enhancement of the work reported in BIB004 , a two-way IoT authentication schemes based on RSA with the use of Trusted Platform Module (TPM). In BIB001 , the authors proposed an authentication protocol based on PUFs that uses the zero-knowledge proof of knowledge (ZKPK) of discrete logarithm. The protocol requires a user to input a password to the device each time it requires authentication. The limitation of this protocol is providing the authentication between the user end device and the server without also taking into consideration authentication between end devices. The use of ZKPK increased the complexity of this protocol. After doing a security analysis for some of previous scheme for the authentication of devices in Low-earth-orbit satellite (LEOs) communication systems, the authors of proposed an authentication and key management scheme based on Elliptic curve and symmetric cryptography. The proposed scheme provides a mutual authentication between the user and the Network Control Center (NCC), besides its resilience to stolen verifier, replay, DoS, password guessing, and impersonation attacks. The authors of BIB016 presented a two-factor device authentication scheme. They used both the digital signature and a novel factor called the device capability. Device capability is similar to a functional operation solved by the device, which could be a mathematical challenge or even a cryptographic-based puzzle. Such scheme can be used to authenticate both the end-device and the server too. Using a secure TLS channel, the device sends a request for communication to the server, the server then sends a nonce encrypted with its private key and the timestamp to avoid replay attacks. The device then decrypts the signature, solves the nonce with functional operation, signs the result with its private key and sends it back to the server. The server then checks the valid signature and the result of the functional operation to authenticate the device. The authors of proposed a new authentication scheme for IoT system based on PUF. The scheme is based on generating a response of a challenge, then feeding the response to another PUF as a challenge. The two PUFs are connected using Linear Feedback Shift Register (LFSR). The main drawback of such design is its complexity, and the lack of security analysis. Machine learning attacks to create a model of PUF is not considered. The authors of BIB017 proposed a new authentication scheme for IoT systems based on blockchain called Bubbles-of-Trust. The idea is to divide the devices into virtual zones called bubbles in which they identify and trust each other (concept of grouping). Then, the communication (transaction) between different devices is controlled and validated by the public blockchain implemented using Ethereum. To provide authentication for IoT system integrated in cloud computing environment, the authors of BIB019 provided a lightweight two-factor authentication scheme based on one-way hashing and XOR operation. The authentication process is made up of three steps: registration, verification, and password renewal step. The computation cost of such scheme is considered and its efficiency is shown in resource-constrained environments. To ensure the security of IoT systems, the authors of BIB018 took advantage of PKI by using X.509 digital certificates to ensure the authentication of devices within the system. Such certificates can be used for device identification and also the device integrity. The authors of BIB002 proposed a model to addresses the IoT security issues. They used Object Naming service (ONS) to query the Domain Name Server (DNS) to search for information about a device using either its IP address or electronic product Codes (EPC). They ensured the anonymity of the requester by asking multiple ONSs to hide the identity and using multiple encryption layers using the encryption key of the router during transmission. Table 2 summarizes the different surveyed IoT authentication schemes with a classification based on the taxonomy presented in Section 4 as well as a summary of their advantages and disadvantages. The following abbreviations are used for the below criteria: • IoT Layer: A, Application; N, Network; P, Perception. • Procedure: 1, One-way; 2, Two-way; 3, Three-way. • Architecture: C, Centralized; D, Distributed; F, Flat; H, Hierarchical. • Hardware-Based: I, Implicit; E, Explicit.

A Survey of Internet of Things (IoT) Authentication Schemes † <s> Related Works <s> In this paper, a comprehensive survey of authentication protocols for Internet of Things (IoT) is presented. Specifically more than forty authentication protocols developed for or applied in the context of the IoT are selected and examined in detail. These protocols are categorized based on the target environment: (1) Machine to Machine Communications (M2M), (2) Internet of Vehicles (IoV), (3) Internet of Energy (IoE), and (4) Internet of Sensors (IoS). Threat models, countermeasures, and formal security verification techniques used in authentication protocols for the IoT are presented. In addition a taxonomy and comparison of authentication protocols that are developed for the IoT in terms of network model, specific security goals, main processes, computation complexity, and communication overhead are provided. Based on the current survey, open issues are identified and future research directions are proposed. <s> BIB001 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Related Works <s> Internet of Things (IoT) consists of a large number of connected objects that are communicating with each other. In order to support trusted communication between IoT objects, effective authentication procedures should be applied between the communicating entities. In this paper a survey of IoT authentication techniques, which are proposed in the literature, is presented. The survey aims to help other researchers in delving into the details of such techniques by going through their classification and comparison. The classification has been done based on the inherent features of these authentication technique such as being distributed vs. centralized, flat vs. hierarchical, and more others. A comparison between these techniques according to the used evaluation models and their security analysis is illustrated. <s> BIB002 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Related Works <s> The Internet of Things (IoT) is the technology, which forms the foundation of today's smart and connected world. The IoT is a network used to interconnect embedded devices, such as sensors, which are able to generate, communicate and share data with one another. In such an enormous network of interconnected "smart objects", the identification of a specific object poses a fundamental task that influences all other functions of the system, such as its governance, privacy features, access control, overall architecture, etc. This paper presents a critical review of the prominent and recent authentication techniques for IoT objects. This is followed by a discussion of the limitations of current authentication techniques and future research avenues and opportunities. <s> BIB003 </s> A Survey of Internet of Things (IoT) Authentication Schemes † <s> Related Works <s> Health care is one of the primary beneficiaries of the technological revolution created by Internet of Things (IoT). In the implementation of health care with IoT, wireless body area network (WBAN) is a suitable communication tool. That being the case security has been one of the major concerns to efficiently utilize the services of WBAN. The diverse nature of the technologies involved in WBAN, the broadcast nature of wireless networks, and the existence of resource constrained devices are the main challenges to implement heavy security protocols for WBAN. In this paper we develop a risk-based adaptive authentication mechanism which continuously monitors the channel characteristics variation, analyzes a potential risk using naive Bayes machine learning algorithm and performs adaptation of the authentication solution. Our solution validates both the authenticity of the user and the device. In addition we evaluate the resource need of the selected authentication solution and provide an offloading functionality in case of scarce resource to perform the selected protocol. The approach is novel because it defines the whole adaptation process and methods required in each phase of the adaptation. The paper also briefly describes the evaluation case study - Smart Home eHealth. <s> BIB004

Several research works surveyed the IoT authentication solutions. The authors of BIB001 BIB002 , the authors provided a classification and comparison of different authentication protocols for IoT. The authors created two different classifications: the first one based on whether the authentication technique is being distributed or centralized, flat or hierarchical, and the second one based on the characteristics of the authentication process. In , the authors provided an analytical survey of the existing literature. They outlined the potential issues and challenges of authentication in IoT for further research but without providing a comparison of the existing schemes. The authors of provided different authentication methods used within the IoT context with emphasis on lightweight and mutual authentication methods but without classification or comparison. In BIB003 , a review of the prominent and "recent" authentication techniques for IoT systems is provided with an indication of their limitations and future research trends, but without providing a classification. The authors of provided a classification of selected research works addressing security concerns and authentication methods, in addition to their evaluation in terms of advantages and disadvantages. In , the authors mapped the current state of the art of the authentication in an IoT environment, featuring the difficulties and the primary methods utilized in authentication scheme but without performing any comparison. The authors of BIB004 provided a brief overview of authentication mechanisms with an evaluation of the proposed schemes in the literature. They also evaluated each method based on its resource consumption (e.g., energy, memory, computation and communication).

3D Scanning: A Comprehensive Survey <s> I. INTRODUCTION <s> Contour moiré pictures with good contrast of a full size living body are now available. Construction problems, performance, and accuracy of large instruments are experimentally studied. Techniques to minimize the drawbacks of the moiré contouring are developed. The system has been proved to be ready for practical application to full size living bodies. <s> BIB001 </s> 3D Scanning: A Comprehensive Survey <s> I. INTRODUCTION <s> Blocking activity with intraocular tetrodotoxin prevents the sharpening of the retinotectal map formed during regeneration of the optic nerve. If (under normal conditions) the initially diffuse map sharpens because of correlated activity in neighboring but not distant ganglion cells, then sharpening should also be prevented merely by disrupting the spatiotemporal correlation in the pattern of activity. To test this idea, fish were exposed during regeneration to stroboscopic illumination in a featureless environment, or were maintained in complete darkness. The regenerating cells remained visually responsive after axotomy, and the xenon strobe effectively drove each ganglion cell at a constant latency. The maps formed in the strobe-reared fish were normally oriented, but the multiunit receptive fields were greatly enlarged, averaging 32 degrees. In control regenerates, multiunit receptive fields averaged only 11-12 degrees, nearly the same as for single units. Dark rearing, which allows only spontaneous activity, also resulted in enlarged multiunit receptive fields, averaging more than 28 degrees. Both effects parallel those reported previously with tetrodotoxin block. The mature projection did not become diffuse as a result of the strobe rearing, and the sensitive period corresponded to the early stage of synaptogenesis (20-34 days). Periods of normal visual exposure after 35 days produced very little sharpening of the diffuse maps produced during either strobe or dark rearing. The results are attributed to an activity-dependent stabilization of developing synapses. The correlated firing of neighboring ganglion cells could allow postsynaptic summation of their responses, and the retention of those more effective, retinotopically placed synapses might then occur via a Hebbian mechanism. <s> BIB002 </s> 3D Scanning: A Comprehensive Survey <s> I. INTRODUCTION <s> Historically, the concept of replacing the photographic emulsion by video recording and display occured to several independent groups almost simultanously. Macovsky(1) in USA, Schwomma(2) in Austria and Kopf(3) in West-Germany considered the technique to be based on pure holography. The Lougborough group in England, headed by Butters and Leendertz(4) viewed the technique more as an off-spring of their speckle work and they also introduced the name ESPI as an abbreviation for electronic speckle pattern interferometry. Although the ESPI-name might be considered appropriately descriptive, its association with coarse, annoying speckles has undoubtedly hampered the general acceptance of the technique. Video holography might be a name with better sales appeal, but might be confused with techniques for direct transmission of three-dimensional holographic images(5). <s> BIB003 </s> 3D Scanning: A Comprehensive Survey <s> I. INTRODUCTION <s> This paper presents an introduction to ionic polymer-metal composites and some mathematical modeling pertaining to them. It further discusses a number of recent findings in connection with ion-exchange polymer-metal composites (IPMCs) as biomimetic sensors and actuators. Strips of these composites can undergo large bending and flapping displacement if an electric field is imposed across their thickness. Thus, in this sense they are large motion actuators. Conversely by bending the composite strip, either quasi-statically or dynamically, a voltage is produced across the thickness of the strip. Thus, they are also large motion sensors. The output voltage can be calibrated for a standard size sensor and correlated to the applied loads or stresses. They can be manufactured and cut in any size and shape. In this paper first the sensing capability of these materials is reported. The preliminary results show the existence of a linear relationship between the output voltage and the imposed displacement for almost all cases. Furthermore, the ability of these IPMCs as large motion actuators and robotic manipulators is presented. Several muscle configurations are constructed to demonstrate the capabilities of these IPMC actuators. This paper further identifies key parameters involving the vibrational and resonance characteristics of sensors and actuators made with IPMCs. When the applied signal frequency varies, so does the displacement up to a critical frequency called the resonant frequency where maximum deformation is observed, beyond which the actuator response is diminished. A data acquisition system was used to measure the parameters involved and record the results in real time basis. Also the load characterizations of the IPMCs were measured and it was shown that these actuators exhibit good force to weight characteristics in the presence of low applied voltages. Finally reported are the cryogenic properties of these muscles for potential utilization in an outer space environment of a few Torrs and temperatures of the order of -140 degrees Celsius. These muscles are shown to work quite well in such harsh cryogenic environments and thus present a great potential as sensors and actuators that can operate at cryogenic temperatures. <s> BIB004 </s> 3D Scanning: A Comprehensive Survey <s> I. INTRODUCTION <s> The Ball-Pivoting Algorithm (BPA) computes a triangle mesh interpolating a given point cloud. Typically, the points are surface samples acquired with multiple range scans of an object. The principle of the BPA is very simple: Three points form a triangle if a ball of a user-specified radius p touches them without containing any other point. Starting with a seed triangle, the ball pivots around an edge (i.e., it revolves around the edge while keeping in contact with the edge's endpoints) until it touches another point, forming another triangle. The process continues until all reachable edges have been tried, and then starts from another seed triangle, until all points have been considered. The process can then be repeated with a ball of larger radius to handle uneven sampling densities. We applied the BPA to datasets of millions of points representing actual scans of complex 3D objects. The relatively small amount of memory required by the BPA, its time efficiency, and the quality of the results obtained compare favorably with existing techniques. <s> BIB005 </s> 3D Scanning: A Comprehensive Survey <s> I. INTRODUCTION <s> To measure erythrocyte deformability in vitro, we made transparent microchannels on a crystal substrate as a capillary model. We observed axisymmetrically deformed erythrocytes and defined a deformation index directly from individual flowing erythrocytes. By appropriate choice of channel width and erythrocyte velocity, we could observe erythrocytes deforming to a parachute-like shape similar to that occurring in capillaries. The flowing erythrocytes magnified 200-fold through microscopy were recorded with an image-intensified high-speed video camera system. The sensitivity of deformability measurement was confirmed by comparing the deformation index in healthy controls with erythrocytes whose membranes were hardened by glutaraldehyde. We confirmed that the crystal microchannel system is a valuable tool for erythrocyte deformability measurement. Microangiopathy is a characteristic complication of diabetes mellitus. A decrease in erythrocyte deformability may be part of the cause of this complication. In order to identify the difference in erythrocyte deformability between control and diabetic erythrocytes, we measured erythrocyte deformability using transparent crystal microchannels and a high-speed video camera system. The deformability of diabetic erythrocytes was indeed measurably lower than that of erythrocytes in healthy controls. This result suggests that impaired deformability in diabetic erythrocytes can cause altered viscosity and increase the shear stress on the microvessel wall. <s> BIB006 </s> 3D Scanning: A Comprehensive Survey <s> I. INTRODUCTION <s> Inkjet printing is familiar as a method of printing text and images onto porous surfaces. In the last few years it has been used as a free-form fabrication method for building three-dimensional parts and is being explored as a way of printing electrical and optical devices, especially where these involve organic components. Inkjet printers are also being used to produce arrays of proteins and nucleic acids. The need for a versatile inkjet technology for free-forming materials and for multilayer devices raises a number of materials problems that do not apply to conventional printing of images. Higher resolutions will be needed if organic transistors are to be printed. Also, it must be possible to print pinhole-free layers to avoid shorting of devices. Multiple layers must be printed such that they mix and react to form a single material or such that they form discrete unmixed layers. Printing on dense rather than porous substrates will be the norm. This article reviews the range of materials that has been ... <s> BIB007 </s> 3D Scanning: A Comprehensive Survey <s> I. INTRODUCTION <s> Although in recent years the 3D-scanning field has reached a good level of maturity, it is still far from being perceived by common users as a 3D-photography approach, as simple as standard photography is. The main reason for that is that obtaining good 3D models without human intervention is still very hard. In particular, two problems remain open: automatic registration of single shots and planning of the acquisition session. In this paper we address the second issue and propose a solution to improve the coverage of automatically acquired objects. Rather than searching for the next-best-view in order to minimise the number of acquisitions, we propose a simple and easy-to-implement algorithm limiting our scope to closing gaps (i.e. filling unsampled regions) in roughly acquired models. The idea is very simple: detect holes in the current model and cluster their estimated normals in order to determine new views. Some results are shown to support our approach. <s> BIB008 </s> 3D Scanning: A Comprehensive Survey <s> I. INTRODUCTION <s> The book covers the field of close-range photogrammetry which has diverse fields of application such as manufacturing engineering (e.g. shape verification and reverse engineering), and medicine and entertainment (motion capture and analysis). It provides detailed overviews of analytical methods (e.g. bundle adjustment and image processing), commercial photogrammetric systems and applications. <s> BIB009 </s> 3D Scanning: A Comprehensive Survey <s> I. INTRODUCTION <s> Abstract Progress reporting is an essential management function for successful delivery of construction projects. It relies on tangible data collected from construction job sites, which is then used to compare actual work performed to that planned. One method used to collect actual work data is 3D laser scanning, where the construction site is scanned at different times to generate data, which can then be used to estimate the quantities of work performed within the time interval considered between two successive scans. Photogrammetry is another method for data collection where the geometrical properties of an object on site are generated from its photo image. This paper presents a method, which integrates 3D scanning and photogrammetry in an effort to enhance the speed and accuracy of data collection from construction sites to support progress measurement and project control. The application of the proposed method is demonstrated using a building presently under construction. <s> BIB010 </s> 3D Scanning: A Comprehensive Survey <s> I. INTRODUCTION <s> Structure from motion (SFM) algorithms are known for their ability to reconstruct a sparse point cloud of scenes that were imaged by a series of overlapping photographs. When complemented by stereo-matching algorithms, detailed three-dimensional models can be built from such photograph collections in a fully automated way. Since 2010, a computer vision software package called PhotoScan has been available from the Russian manufacturer AgiSoft LLC. Even though the programme has a straightforward and simple interface, state-of-the-art SFM and other computer vision algorithms are implemented that enable PhotoScan to generate very dense and accurate three-dimensional meshes of various (archaeological) settings captured in two-dimensional imagery. Using a mixture of archaeological case studies, it will be shown that this low-cost application produces excellent results from aerial imagery in a minimum of time with almost no effort needed. Because of its solid and uncomplicated operation, it looks like this application could become an essential part of the aerial archaeologist's toolkit. Copyright © 2011 John Wiley & Sons, Ltd. <s> BIB011 </s> 3D Scanning: A Comprehensive Survey <s> I. INTRODUCTION <s> The importance of landscape and heritage recording and documentation with optical remote sensing sensors is well recognized at international level. The continuous development of new sensors, data capture methodologies and multi-resolution 3D representations, contributes significantly to the digital 3D documentation, mapping, conservation and representation of landscapes and heritages and to the growth of research in this field. This article reviews the actual optical 3D measurement sensors and 3D modeling techniques, with their limitations and potentialities, requirements and specifications. Examples of 3D surveying and modeling of heritage sites and objects are also shown throughout the paper. <s> BIB012 </s> 3D Scanning: A Comprehensive Survey <s> I. INTRODUCTION <s> UAV platforms are nowadays a valuable source of data for inspection, surveillance, mapping and 3D modeling issues. New applications in the short- and close-range domain are introduced, being the UAVs a low-cost alternatives to the classical manned a erial photogrammetry. Rotary or fixed wing UAVs, capable of performing the photogrammetric data acquisition with amateur or SLR digital cameras, can fly in manual, semi-automated and autonomous modes. With a typical photogrammetric pipeline, 3D results like DSM/DTM, contour lines, textured 3D models, vector data, etc. can be produced, in a reasonable automated way. The paper reports the latest developments of UAV image processing methods for photogrammetric applications, mapping and 3D modeling issues. Automation is nowadays necessary and feasible at the image orientation, DSM generation and orthophoto production stages, while accurate feature extraction is still an interactive procedure. New perspectives are also addressed. <s> BIB013 </s> 3D Scanning: A Comprehensive Survey <s> I. INTRODUCTION <s> Dielectric elastomer (DE) actuators are popularly referred to as artificial muscles because their impressive actuation strain and speed, low density, compliant nature, and silent operation capture many of the desirable physical properties of muscle. Unlike conventional robots and machines, whose mechanisms and drive systems rapidly become very complex as the number of degrees of freedom increases, groups of DE artificial muscles have the potential to generate rich motions combining many translational and rotational degrees of freedom. These artificial muscle systems can mimic the agonist-antagonist approach found in nature, so that active expansion of one artificial muscle is taken up by passive contraction in the other. They can also vary their stiffness. In addition, they have the ability to produce electricity from movement. But departing from the high stiffness paradigm of electromagnetic motors and gearboxes leads to new control challenges, and for soft machines to be truly dexterous like their biolo... <s> BIB014 </s> 3D Scanning: A Comprehensive Survey <s> I. INTRODUCTION <s> This paper proposes a novel phase-shifting method for fast, accurate, and unambiguous 3D shape measurement. The basic idea is embedding a speckle-like signal in three sinusoidal fringe patterns to eliminate the phase ambiguity, but without reducing the fringe amplitude or frequency. The absolute depth is then recovered through a robust region-wise voting strategy relying on the embedded signal. Using the theoretical minimum of only three images, the proposed method greatly facilitates the application of phase shifting in time-critical conditions. Moreover, the proposed method is resistant to the global illumination effects, as the fringe patterns used are with a single high frequency. Based on the proposed method, we further demonstrate a real-time, high-precision 3D scanning system with an off-the-shelf projector and a commodity camera. <s> BIB015 </s> 3D Scanning: A Comprehensive Survey <s> I. INTRODUCTION <s> Abstract. The paper presents a home-made 3d scanner, consisting of off-the-shelf components: a camera and a projector. It is intended for monitoring dynamics of riverbed mophology observed under laboratory conditions in a flume, which is currently under construction. Special attention is paid to satisfying high requirements concerning accuracy and precision despite a compact and versatile setup of the system. Preliminary results are shown. <s> BIB016 </s> 3D Scanning: A Comprehensive Survey <s> I. INTRODUCTION <s> Abstract Image-based modeling is an appropriate technique to create 3D models of cultural heritage objects, which starts with the basic task of designing the camera network. This task is, however, quite crucial in practical applications because it needs a thorough planning and a certain level of experience. The optimal camera network is designed when certain accuracy demands are fulfilled with a reasonable effort, namely keeping the number of camera shots at a minimum. In this study, we report on the development of an automated method for designing the optimal camera network for a given cultural heritage building or statue. Starting from a rough point cloud derived from a video image stream, the initial configuration of the camera network is designed, assuming a high-resolution HR state-of-the-art non-metric camera. To improve the image coverage and accuracy, we use a mathematical non-linear optimization with constraints. Furthermore, synthetic images are created to guide the camera operator to the designed images. From the first experimental test, we found that a target accuracy of 10 mm could be maintained although the initial number of more than 300 high-resolution images got reduced to less than 90 for the final, optimized network. <s> BIB017 </s> 3D Scanning: A Comprehensive Survey <s> I. INTRODUCTION <s> An algorithm for the reconstruction of the 3D shape of the surface of a micro-object from a stereo pair of images obtained on a raster electron microscope (REM) has been considered. A model of building an image in REM has been presented. The SIFT algorithm was used for determination of the correspondence points. The correspondence points are used to determine the mutual position of the object in a stereo pair by the RANSAC method. A set of points is created in the 3D space, which is later interpolated to reconstruct the 3D surface. <s> BIB018 </s> 3D Scanning: A Comprehensive Survey <s> I. INTRODUCTION <s> Abstract This paper presents a generative statistical approach to automatic 3D building roof reconstruction from airborne laser scanning point clouds. In previous works, bottom-up methods, e.g., points clustering, plane detection, and contour extraction, are widely used. Due to the data artefacts caused by tree clutter, reflection from windows, water features, etc., the bottom-up reconstruction in urban areas may suffer from a number of incomplete or irregular roof parts. Manually given geometric constraints are usually needed to ensure plausible results. In this work we propose an automatic process with emphasis on top-down approaches. The input point cloud is firstly pre-segmented into subzones containing a limited number of buildings to reduce the computational complexity for large urban scenes. For the building extraction and reconstruction in the subzones we propose a pure top-down statistical scheme, in which the bottom-up efforts or additional data like building footprints are no more required. Based on a predefined primitive library we conduct a generative modeling to reconstruct roof models that fit the data. Primitives are assembled into an entire roof with given rules of combination and merging. Overlaps of primitives are allowed in the assembly. The selection of roof primitives, as well as the sampling of their parameters, is driven by a variant of Markov Chain Monte Carlo technique with specified jump mechanism. Experiments are performed on data-sets of different building types (from simple houses, high-rise buildings to combined building groups) and resolutions. The results show robustness despite the data artefacts mentioned above and plausibility in reconstruction. <s> BIB019 </s> 3D Scanning: A Comprehensive Survey <s> I. INTRODUCTION <s> The present paper compares between four data sources for creating digital terrain models (DTMs), based on analytical aerial photogrammetry, laser scanning, total station (TS) and global positioning system surveys. The case study presents the comparative results obtained using the methods listed above. After obtaining the ground coordinates of points using each method, DTMs were created using Surfer Software to establish comparison consistency. To check accuracy of each method used, a series of 100 independent points were collected at random positions over the case study area using the TS for horizontal position and a digital level for elevation. The comparison of the results of the methods was performed based on accuracy and efficiency aspects. The results have shown that TS and terrestrial laser scanners (TLSs) could represent precise, effective and rapid solutions to produce economical and accurate DTMs. Also, using TLSs allow more various possibilities of data representation. <s> BIB020 </s> 3D Scanning: A Comprehensive Survey <s> I. INTRODUCTION <s> The ongoing development of advanced techniques in photogrammetry, computer vision (CV), robotics and laser scanning to efficiently acquire three dimensional geometric data offer new possibilities for many applications. The output of these techniques in the digital form is often a sparse or dense point cloud describing the 3D shape of an object. Viewing these point clouds in a computerized digital environment holds a difficulty in displaying the visible points of the object from a given viewpoint rather than the hidden points. This visibility problem is a major computer graphics topic and has been solved previously by using different mathematical techniques. However, to our knowledge, there is no study of presenting the different visibility analysis methods of point clouds from a photogrammetric viewpoint. The visibility approaches, which are surface based or voxel based, and the hidden point removal (HPR) will be presented. Three different problems in close range photogrammetry are presented: camera network design, guidance with synthetic images and the gap detection in a point cloud. The latter one introduces also a new concept of gap classification. Every problem utilizes a different visibility technique to show the valuable effect of visibility analysis on the final solution. <s> BIB021 </s> 3D Scanning: A Comprehensive Survey <s> I. INTRODUCTION <s> The vast amount of public photographic data posted and shared on Facebook, Instragram and other forms of social media offers an unprecedented visual archive of the world. This archive captures events ranging from birthdays, trips, and graduations to lethal conflicts and human rights violations. Because this data is public, it has led to a new genre of journalism, one led by citizens finding, analyzing, and synthesizing data into stories that describe important events. To support this, we have built a set of browser-based tools for the calibration and validation of online images. This paper presents these tools in the context of their use in finding two separate lost burial locations. Often, these locations would have been marked with a headstone or tomb, but for the very poor, the forgotten, or the victims of extremist violence buried in unmarked graves, the geometric cues present in a photograph may contain the most reliable information about the burial location. The tools described in this paper allow individuals without any significant geometry background to utilize those cues to locate these lost graves, or any other outdoor image with sufficient correspondences to the physical world. We highlight the difficulties that arise due to geometric inconsistencies between corresponding points, especially when significant changes have occurred in the physical world since the photo was taken, and visualization features on our browser-based tools that help users address this. <s> BIB022 </s> 3D Scanning: A Comprehensive Survey <s> I. INTRODUCTION <s> Photogrammetry is the earliest technique used to collect data for topographic mapping. The recent development in aerial photogrammetry is the used of large format digital aerial camera for producing topographic map. The aerial photograph can be in the form of metric or non-metric imagery. The cost of mapping using aerial photogrammetry is very expensive. In certain application, there is a need to map small area with limited budget. Due to the development of technology, small format aerial photogrammetry technology has been introduced and offers many advantages. Currently, digital map can be extracted from digital aerial imagery of small format camera mounted on light weight platform such as unmanned aerial vehicle (UAV). This study utilizes UAV system for large scale stream mapping. The first objective of this study is to investigate the use of light weight rotary-wing UAV for stream mapping based on different flying height. Aerial photograph were acquired at 60% forward lap and 30% sidelap specifications. Ground control points and check points were established using Total Station technique. The digital camera attached to the UAV was calibrated and the recovered camera calibration parameters were then used in the digital images processing. The second objective is to determine the accuracy of the photogrammetric output. In this study, the photogrammetric output such as stereomodel in three dimensional (3D), contour lines, digital elevation model (DEM) and orthophoto were produced from a small stream of 200m long and 10m width. The research output is evaluated for planimetry and vertical accuracy using root mean square error (RMSE). Based on the finding, sub-meter accuracy is achieved and the RMSE value decreases as the flying height increases. The difference is relatively small. Finally, this study shows that UAV is very useful platform for obtaining aerial photograph and subsequently used for photogrammetric mapping and other applications. <s> BIB023 </s> 3D Scanning: A Comprehensive Survey <s> I. INTRODUCTION <s> Traditional road surveying methods rely largely on in-situ measurements, which are time consuming and labor intensive. Recent Mobile Laser Scanning (MLS) techniques enable collection of road data at a normal driving speed. However, extracting required information from collected MLS data remains a challenging task. This paper focuses on examining the current status of automated on-road object extraction techniques from 3D MLS points over the last five years. Several kinds of on-road objects are included in this paper: curbs and road surfaces, road markings, pavement cracks, as well as manhole and sewer well covers. We evaluate the extraction techniques according to their method design, degree of automation, precision, and computational efficiency. Given the large volume of MLS data, to date most MLS object extraction techniques are aiming to improve their precision and efficiency. <s> BIB024 </s> 3D Scanning: A Comprehensive Survey <s> I. INTRODUCTION <s> This paper presents a new methodology for the calculation of a digital elevation model (DEM) and the corresponding orthophotography at very high resolution (20 cm for the DEM, 10 cm for the orthophotography) on a large surface (more than 3 km2). Image acquisition was done with an off-the-shelf camera (Sony NEX-5N) attached to a delta kite. Developing the most accessible methodology being one of the goals of this study, the overall cost of the apparatus is limited to less than 1500 euros. This method has been developed to form an alternative with acquisitions from small Unmanned Aerial Vehicles (UAV). In particular, kite as payload carrier is seen as being complementary to lightweight multicopters, which are increasingly used for remote sensing applications [1, 2]. Kite is of a strong interest when the use of radio-controlled and motorized systems is hampered by local regulation and/or when too strong winds keep these lightweight devices on ground. <s> BIB025 </s> 3D Scanning: A Comprehensive Survey <s> I. INTRODUCTION <s> ABSTRACT Icing thickness parameter is the basic data for power sector to make decision for icing accident prevention. In this paper, a transmission line icing measurement method is proposed. It used the photogrammetry method to realize icingparameters measurement through the integration of high reso lution camera, laser range finder and inertial measurement unit. Compared with traditional icing measur ement method, this method is flexible and is the effective supplem ent of thefixed icing detection terminal. And its high accuracy measurement guarantees the reliability of the icing thicknessparame ters.Keywords: Icing thickness, parameter measurement, multi-sensor integration, photogrammetry 1. INTRODUCTION Frequent accidents of transmission line icing happened in disaster weather such as snowstorm, freezing rain and etc.Freezing rain and icing increase the load of transmission line and cause power line broken, tower (bar) collapse andflashove r. These accidents seriously threaten the safe operation of the power system. <s> BIB026 </s> 3D Scanning: A Comprehensive Survey <s> I. INTRODUCTION <s> Demand for high-volume 3D scanning of real objects is rapidly growing in a wide range of applications, including online retailing, quality-control for manufacturing, stop motion capture for 3D animation, and archaeological documentation and reconstruction. Although mature technologies exist for high-fidelity 3D model acquisition, deploying them at scale continues to require non-trivial manual labor. We describe a system that allows non-expert users to scan large numbers of physical objects within a reasonable amount of time, and with greater ease. Our system uses novel view- and path-planning algorithms to control a structured-light scanner mounted on a calibrated motorized positioning system. We demonstrate the ability of our prototype to safely, robustly, and automatically acquire 3D models for large collections of small objects. <s> BIB027 </s> 3D Scanning: A Comprehensive Survey <s> I. INTRODUCTION <s> Abstract This study proposes an efficient Bundle Adjustment (BA) model for oblique aerial photogrammetry to reduce the number of unknown parameters and the dimensions of a non-linear optimization problem. Instead of serving as independent exterior orientations, oblique camera poses are parameterized with nadir camera poses and constant relative poses between oblique and nadir cameras. New observation functions are created with image points as a function of the nadir pose and the relative pose parameters. With these observation functions, the problem of BA is defined as finding optimal unknown parameters by minimizing the total difference between estimated and observed image points. A Gauss-Newton optimization method is utilized to provide a solution for this least-square problem with a reduced normal equation, which plays a very critical role in the convergence and efficiency of BA. Compared with traditional BA methods, the number of unknown parameters and the dimension of the normal equations decrease, this approach dramatically reduces the computational complexity and memory cost especially for large-scale scenarios with a number of oblique images. Four synthetic datasets and a real dataset were used to check the validation and the accuracy of the proposed method. The accuracy of the proposed method is very close to that of the traditional BA method, but the efficiency can be significantly improved by the proposed method. For very large-scale scenarios, the proposed method can still address the limitation of memory and orientate all images captured by an oblique aerial multi-camera system. <s> BIB028 </s> 3D Scanning: A Comprehensive Survey <s> I. INTRODUCTION <s> In this paper the development of a wearable 3D FMCW radar sensor system for firefighters to enhance eyesight in smoky surroundings during hazardous situations is described. In order to obtain a real-time capable, small and lightweight sensor, two orthogonal oriented frequency-scanning meander-line antennas are combined with small broadband 57 GHz to 63 GHz FMCW radar-modules. This arrangement is used to scan both directions, azimuth and elevation, in a fast time-multiplexed mode. The gathered time-domain datasets for both scanning directions are transferred to 2D projections. A following data-processing step merges both of these orthogonal images into a 3D-image of the monitored area. As no mechanical parts are involved, this very simple 2 Channel radar concept is suitable for a high degree of miniaturization. Integration of the whole system into a wearable system deployable for a broad field of applications in security and defense seems to be feasible. <s> BIB029 </s> 3D Scanning: A Comprehensive Survey <s> I. INTRODUCTION <s> The use of virtual reality (VR) has been exponentially increasing and due to that many researchers have started to work on developing ::: new VR based social media. For this purpose it is important to have an avatar of the user which look like them to be easily generated by the ::: devices which are accessible, such as mobile phones. In this paper, we propose a novel method of recreating a 3D human face model captured ::: with a phone camera image or video data. The method focuses more on model shape than texture in order to make the face recognizable. We ::: detect 68 facial feature points and use them to separate a face into four regions. For each area the best fitting models are found and are further ::: morphed combined to find the best fitting models for each area. These are then combined and further morphed in order to restore the original ::: facial proportions. We also present a method of texturing the resulting model, where the aforementioned feature points are used to generate ::: a texture for the resulting model. <s> BIB030 </s> 3D Scanning: A Comprehensive Survey <s> I. INTRODUCTION <s> High-precision and high-resolution topography is the basis of the quantitative study of active faults. Light detection and ranging lidar is currently the most popular method for obtaining such data, but its relatively high cost greatly limits its use in many geoscience applications. Recently, with the rapid development of computer vision science and the growing application of small unmanned aerial vehicles UAVs, Structure from Motion SfM photogrammetry shows great potential for providing topographic information of comparable resolution and precision to lidar surveys, but at significantly lower cost. In this study, we examined the applicability of SfM photogrammetry in modelling the topography of the fault zone using images acquired with a low-cost digital camera mounted on a UAV over the Haiyuan fault. The resolution and accuracy of the SfM-derived topography were evaluated in detail using existing airborne lidar data as a benchmark. The results show that the density of the point cloud generated by SfM photogrammetry is nearly 70 times higher than that from the airborne lidar. Furthermore, considering the errors in the lidar data itself, the precision of the SfM point cloud is comparable to that of the lidar point cloud. Overall, our results demonstrate that the UAV-based SfM photogrammetry method can provide an inexpensive, effective, and flexible alternative to airborne lidar for the topography mapping of the fault zone. <s> BIB031 </s> 3D Scanning: A Comprehensive Survey <s> I. INTRODUCTION <s> This paper proposes a new oil painting reproduction method using 3D printing to compensate for the deficiencies of the existing methods. First, 3D reconstruction of oil paintings is completed by photogrammetry; the oil painting color and the 3D geometric information are recovered better by acquiring several sets of orthophotomaps, and modeling accuracy is ensured with a control mesh or by flattening. Next, the contours and hypsometric tints of the 3D model for oil paintings are generated using contour tracing algorithm, and the image segmentation of renderings is completed using RGB image segmentation algorithm, with the layered section extracted from each layer and the 3D geometric information converted into 2D plane information. Finally, the 3D models of oil paintings are presented through UV inkjet printing with images superimposed layer upon layer, and stereoscopic reproduction of oil paintings is completed based on the orthophotomaps printed from the 3D models. <s> BIB032

Automated scene interpretation is a topic of major interest in photogrammetry, Remote Sensing (RS), and computer vision. 3D shape measurement, also known as range imaging and depth sensing, plays an important role in many applications. However, the mainstream methods do not provide measurements accurate and dense enough for a reliable 3D reconstruction BIB015 - BIB030 . 3D scanning of scenes and multiple objects is time-consuming, and thus bottlenecked by manual labor. Nevertheless, since the associated processing techniques are under constant improvements, the latter drawback has been substantially alleviated in the last few years. It should be noted that usually, when the object of interest has a distinctive surface, photogrammetric techniques are favored. On the other hand, there are not enough free benchmark databases available for testing the associated algorithms and procedures. It should be noted that photogrammetry may not be as efficient as Total Station (TS) and ground-based Interferometric Synthetic-Aperture Radar (InSAR) systems, due to accuracy and response-time requirements. In , a differential 3D scanning procedure was proposed, in each iteration of which, the designer creates a 3D model, prints it using a 3D printer, edits it, and then scans it back to apply the changes. The true changes made to the model are automatically detected based on the reference model, which are chosen based on the coefficients related to the printing accuracy, aiming at minimizing the accumulation of errors. The method does not work well on shiny and translucent objects, and as 3D printers are slow, this type of iterative M. Daneshmand, A. Helmi, E. Avots, F. Noroozi, H. S. Arslan, J. Gorbova, R. E. Haamer, and G. Anbarjafari are with iCV Research Lab, Institute of Technology, University of Tartu, Tartu, Estonia. e-mail: [email protected] F. Alisinanoglu and G. Anbarjafari are also with Department of Electrical and Electronic Engineering, Hasan Kalyoncu University, Gaziantep, Turkey. C. Ozcinar is with Trinity College Dublin (TCD), Dublin 2, Ireland. e-mail: [email protected] workflow is limited by the number of iterations. In BIB024 , methods of 3D on-road object extraction from Mobile Laser Scanning (MLS) data were represented, e.g. for road surface, pavement cracks, road markings and manhole covers, where the samples were collected at driving speeds. In BIB020 , a comparison of analytical aerial photogrammetry, i.e. B/W stereo photographs, Time-Of-Flight (TOF) LS-based measurement mechanic, TS based on ground control points and rover with GPS surveys was provided in terms of generating Digital Terrain Models (DTM). Reportedly, TS led to the best accuracy, although demonstrated a very slow performance in data collection, with LS being second, which demands complex data analyses. Aerial photogrammetry required expensive equipment to be viable, and Real-Time Kinematic (RTK) GPS was the most unreliable one. In BIB027 , an easy-to-use motorized and automated system was developed that scans 3D scenes in a reasonable time frame, which can be utilized for viewing scalable multi-object scenes, as well as for path planning purposes. The resolution they achieved was 0.1 mm on a variety of different objects. Due to the fact that 3D scanning with Light Detection And Ranging (LiDAR) is expensive, in BIB031 , it was attempted to achieve comparable results on topography mapping of a fault zone with a low-cost digital camera mounted on a UAV. They used the Agisoft PhotoScan software BIB011 for processing the images in order to reconstruct the ground. Comparisons with LiDAR showed nearly no difference in accuracy, but the Structure-From-Motion (SFM)-based photogrammetry method yielded a denser point cloud. In BIB013 , an evaluation of Unmanned Aerial Vehicle (UAV)-and photogrammetry-based 3D scanning methods was provided. It revealed that UAVs have a lower operational cost, where the task can usually be performed by two operators, but often carry small or medium format cameras, which is why they require more images to reach the same accuracy as conventional methods. In BIB028 , an Reduced Bundle Adjustment (RBA) method was proposed for reducing the computational complexity and memory cost of the task of Bundle Adjustment (BA). It uses both oblique and nadir camera data to fixate their orientations and positions. The accuracy of the proposed method is slightly poor, but the efficiency is significantly improved. One of the important applications of RS is in exploring, recording or documenting objects, landscapes and heritage sites BIB012 . New sensors and multi-resolution 3D scanning and visualization methodologies developed in the foregoing context are aimed at purposes such as mapping and conservation. One of the applications of PaRS is in improving the speed and accuracy of progress reporting, where a construction site is scanned regularly, and the scans are compared to each other, as well as to the planned outcome, in order to measure the amount of work done quantitatively, and help make management and control decisions BIB010 . A homemade 3D scanner consisting of a camera and a projector was built in BIB016 , aiming at relatively high levels of accuracy and precision, as well as a compact and dexterous setup. As a case-study, the system was utilized for monitoring a flume under construction. As demonstrated in , 3D image correlation photogrammetry, possibly in combination with stroboscopic illumination BIB002 or high-speed video cameras BIB006 , , has applications in full-field dynamic displacement, deformation, stress and strain measurement, e.g. for ionic polymeric and dielectric elastomer artificial muscles BIB004 , BIB014 , and presents a higher robustness and dynamic range than other full-field imaging approaches such as 3D Electronic Speckle Pattern Interferometry (ESPI) BIB003 or moiré topography BIB001 . In , pulsed laser was used to investigate the behavior of a flywheel in a spin pit at a maximum speed of 35,000 Revolutions per Minute (RpM), and a notched rubber dogbone sample was pulled to failure in 38 ms, where numerous strain maps were obtained. While displaying a point cloud obtained through photogrammetry, it should be decided which points are visible from a given viewpoint, where surface-and voxel-based methods BIB005 , have been proposed for Hidden Point Removal (HPR), which have been investigated in BIB021 , along with issues such as guidance with synthetic images BIB005 , camera network design BIB017 , BIB009 and gap detection and classification BIB008 . Photogrammetry can make use of the geometric data present in data resources shared through social media such as Facebook and Instagram, in order to calibrate and recover information about important facts, events and locations, such as burial places, with possible changes occurring from the time of taking an image to another one potentially introducing further challenges BIB022 . In BIB029 , a compact 3D imaging radar based on FrequencyModulated Continuous-Wave (FMCW)-driven frequencyscanning antennas was proposed for a small and lightweight radar-system, which is equipped for catching two orthogonal 2D pictures. Angular and parallel determination of these pictures are adequate to supply vital data about the field in basic circumstances. Moreover, a fundamental algorithmic approach was proposed for joining the two pictures to a pseudo 3D model of the landscape. As the compass time per check is under 1 ms, and every 3D picture is developed from only two filters, the frame-rates achievable with this framework design are appropriate for real-time utilization. In BIB018 , an algorithm has been proposed that makes it conceivable to fabricate a 3D model of a miniaturized object on the premise of a stereo combination of pictures acquired by a REM. The parallax in the stereo match is required for revising the performance of the algorithm. The basic grid is computed through Random Sample Consensus (RANSAC). The algorithm is adapted for work with vast magnification coefficients, and can utilize both parallel and point-of-view projection models. In BIB019 , a hybrid generative statistical framework was proposed for automatic extraction and reconstruction of building roofs from ALS point clouds. In BIB025 , a method was proposed which allows to obtain DEMs and the corresponding orthophotography of surfaces larger than 3 km 2 at VHR. They used two delta kites of 4 and 10 km 2 and a SONY NEX-5N camera BIB023 , which met a compromise between weight, image quality and geometry, and cost. In the proposed method, the kit flight angle stability was properly estimated by taking several flights with different conditions and flight-line lengths. One of the limitations of the proposed method is that the route of flight should be open and clear. The importance of icing thickness measurements of transmission lines for preventing possible resulting accidents has been investigated in BIB026 through photogrammetry. The proposed system consists of a High-Resolution (HR) camera, a laser range finder and an inertial measurement unit, which has led to a flexible and accurate performance along with a conventional fixed terminal. In BIB032 , photogrammetry was used for 3D reconstruction of oil paintings. They obtained several sets of orthophotomaps to create 3D models of oil paintings through UV inkjet printing BIB007 . They used the collinearity equation, i.e. the condition equation, in order to represent the image details, followed by reconstructive accuracy analysis. The contours and hypsometric tints of the model were determined by using spot color swatches in four colors for a UV printer, in order to print 1-10-layer spot color swatches. Then RGB image segmentation and section extraction were performed. The results showed that digital photogrammetry can effectively establish a colorful 3D model for oil paintings. Coastal monitoring is fundamental to studying dune and beach behavior related to natural and anthropogenic factors, as well as coastal management programs. Various tools have been applied for the foregoing task, including LiDAR, satellite images and TLS. This paper reviews various 3D scanning technologies, including close-range, aerial, SFM and terrestrial photogrammetry, along with mobile, terrestrial and airborne LS, followed by TOF, structured-light and phase-comparison approaches, and some of the existing comparative and combinational studies. Last but not least, outlier detection and surface fitting processes will be briefly discussed as well. The purpose of this paper is to help grasp an understanding of the state-of-the-art in the area of 3D scanning. Therefore, the articles to be reviewed have been picked up from the list of publications by the most highly accredited conferences and journals in the last four years up to the time the paper has been written, as well as a few older, but most frequently cited, seminal, ones.

3D Scanning: A Comprehensive Survey <s> C. Relative Pose Measurement of Satellite and Rocket <s> Determining the rigid transformation relating 2D images to known 3D geometry is a classical problem in photogrammetry and computer vision. Heretofore, the best methods for solving the problem have relied on iterative optimization methods which cannot be proven to converge and/or which do not effectively account for the orthonormal structure of rotation matrices. We show that the pose estimation problem can be formulated as that of minimizing an error metric based on collinearity in object (as opposed to image) space. Using object space collinearity error, we derive an iterative algorithm which directly computes orthogonal rotation matrices and which is globally convergent. Experimentally, we show that the method is computationally efficient, that it is no less accurate than the best currently employed optimization methods, and that it outperforms all tested methods in robustness to outliers. <s> BIB001 </s> 3D Scanning: A Comprehensive Survey <s> C. Relative Pose Measurement of Satellite and Rocket <s> This paper presents a method of measuring the relative pose of Satellite-rocket separation based on image measurement. In the camera parameters are known, the camera installed on the rocket launcher, and more than six cooperation signs are fixed on the satellite. Using cameras take pictures of the cooperation signs in real time. Based on the basic principle of monocular vision pose estimation. The relative pose relationship between satellite and rocket is solved by combining EPnP with Orthogonal Iteration. The simulation experiment and the semi - physical simulation experiment prove the feasibility and accuracy of the method. The measurement error is calculated according to the error calculation method proposed in this paper. The results show that the relative position measurement error and the relative attitude measurement error are within the measurement range. <s> BIB002

Image-based measurements can be used for calculating the relative pose of satellite-rocket separations BIB002 . In the case the camera parameters are known, the camera has been installed on the rocket launcher, and more than six cooperation signs have been set on the satellite, the latter are captured in realtime, and the pose estimation task is handled by combined virtues of the Efficient Perspective-n-Point (EPnP) and Orthogonal Iteration (OI) BIB001 methods, the latter being referred to as LHM as well, which is abbreviated from the last names of the authors, i.e. Lu, Hager and Mjolsness. In the aforementioned study, an accuracy assessment framework has been proposed as well, which has verified the reliability of the foregoing strategy in terms of estimating the relative position and attitude. III. CLOSE-RANGE PHOTOGRAMMETRY (CRP) CRP is one of the most prevalent variants of photogrammetry, whose applications will be discussed in this section.

3D Scanning: A Comprehensive Survey <s> A. Automatic Camera Calibration <s> Automatic camera calibration via self-calibration with the aid of coded targets is now very much the norm in closerange photogrammetry. This is irrespective of whether the cameras to be calibrated are high-end metric, or the digital SLRs and consumer-grade models that are increasingly being employed for image-based 3D measurement. Automation has greatly simplified the calibration task, but there are real prospects that important camera calibration issues may be overlooked in what has become an almost black-box operation. This paper discusses the impact of a number of such issues, some of which relate to the functional model adopted for self-calibration, and others to practical aspects which need to be taken into account when pursuing optimal calibration accuracy and integrity. Issues discussed include interior orientation stability, calibration reliability, focal plane distortion, image point distribution, variation in lens distortion with image scale, colour imagery and chromatic aberration, and whether 3D object space control is warranted. By appreciating and accounting for these issues, users of automatic camera calibration will enhance the prospect of achieving an optimal recovery of scene-independent camera calibration parameters. <s> BIB001 </s> 3D Scanning: A Comprehensive Survey <s> A. Automatic Camera Calibration <s> Abstract Metric calibration is a critical prerequisite to the application of modern, mostly consumer-grade digital cameras for close-range photogrammetric measurement. This paper reviews aspects of sensor modelling and photogrammetric calibration, with attention being focussed on techniques of automated self-calibration. Following an initial overview of the history and the state of the art, selected topics of current interest within calibration for close-range photogrammetry are addressed. These include sensor modelling, with standard, extended and generic calibration models being summarised, along with non-traditional camera systems. Self-calibration via both targeted planar arrays and targetless scenes amenable to SfM-based exterior orientation are then discussed, after which aspects of calibration and measurement accuracy are covered. Whereas camera self-calibration is largely a mature technology, there is always scope for additional research to enhance the models and processes employed with the many camera systems nowadays utilised in close-range photogrammetry. <s> BIB002

Obtaining CRP measurements requires calibration of the digital camera based on sensor modeling, which can be achieved through self-calibration. It has been reviewed based on targeted planar arrays and target-less scenes in BIB002 . In a CRP-image-based 3D measurement context, the camera calibration parameters for DSLR cameras , as well as highend or consumer-grade ones, need to be calculated independently from the characteristics of the scene at hand, which can be performed automatically based on coded targets BIB001 . However, the latter study suggests that a set of considerations are required for an optimally accurate and reliable calibration procedure, which concern either the self-calibration functional model or practical issues, and among others, include the effect of image scales on lens distortions, color imaging and chromatic calibration, focal plane distortion, 3D object space control, image point distribution and interior orientation stability.