Datasets:

sswt

/

arxiv_sample_50K

like 0

MultiQA An Empirical Investigation of Generalization and Transfer in Reading Comprehension ; A large number of reading comprehension RC datasets has been created recently, but little analysis has been done on whether they generalize to one another, and the extent to which existing datasets can be leveraged for improving performance on new ones. In this paper, we conduct such an investigation over ten RC datasets, training on one or more source RC datasets, and evaluating generalization, as well as transfer to a target RC dataset. We analyze the factors that contribute to generalization, and show that training on a source RC dataset and transferring to a target dataset substantially improves performance, even in the presence of powerful contextual representations from BERT Devlin et al., 2019. We also find that training on multiple source RC datasets leads to robust generalization and transfer, and can reduce the cost of example collection for a new RC dataset. Following our analysis, we propose MultiQA, a BERTbased model, trained on multiple RC datasets, which leads to stateoftheart performance on five RC datasets. We share our infrastructure for the benefit of the research community.

Reliable Fidelity and Diversity Metrics for Generative Models ; Devising indicative evaluation metrics for the image generation task remains an open problem. The most widely used metric for measuring the similarity between real and generated images has been the Fr'echet Inception Distance FID score. Because it does not differentiate the fidelity and diversity aspects of the generated images, recent papers have introduced variants of precision and recall metrics to diagnose those properties separately. In this paper, we show that even the latest version of the precision and recall metrics are not reliable yet. For example, they fail to detect the match between two identical distributions, they are not robust against outliers, and the evaluation hyperparameters are selected arbitrarily. We propose density and coverage metrics that solve the above issues. We analytically and experimentally show that density and coverage provide more interpretable and reliable signals for practitioners than the existing metrics. Code httpsgithub.comclovaaigenerativeevaluationprdc.

Variance Loss in Variational Autoencoders ; In this article, we highlight what appears to be major issue of Variational Autoencoders, evinced from an extensive experimentation with different network architectures and datasets the variance of generated data is significantly lower than that of training data. Since generative models are usually evaluated with metrics such as the Frechet Inception Distance FID that compare the distributions of features of real versus generated images, the variance loss typically results in degraded scores. This problem is particularly relevant in a two stage setting, where we use a second VAE to sample in the latent space of the first VAE. The minor variance creates a mismatch between the actual distribution of latent variables and those generated by the second VAE, that hinders the beneficial effects of the second stage. Renormalizing the output of the second VAE towards the expected normal spherical distribution, we obtain a sudden burst in the quality of generated samples, as also testified in terms of FID.

An Optimal Statistical and Computational Framework for Generalized Tensor Estimation ; This paper describes a flexible framework for generalized lowrank tensor estimation problems that includes many important instances arising from applications in computational imaging, genomics, and network analysis. The proposed estimator consists of finding a lowrank tensor fit to the data under generalized parametric models. To overcome the difficulty of nonconvexity in these problems, we introduce a unified approach of projected gradient descent that adapts to the underlying lowrank structure. Under mild conditions on the loss function, we establish both an upper bound on statistical error and the linear rate of computational convergence through a general deterministic analysis. Then we further consider a suite of generalized tensor estimation problems, including subGaussian tensor PCA, tensor regression, and Poisson and binomial tensor PCA. We prove that the proposed algorithm achieves the minimax optimal rate of convergence in estimation error. Finally, we demonstrate the superiority of the proposed framework via extensive experiments on both simulated and real data.

Learning to Shadow Handdrawn Sketches ; We present a fully automatic method to generate detailed and accurate artistic shadows from pairs of line drawing sketches and lighting directions. We also contribute a new dataset of one thousand examples of pairs of line drawings and shadows that are tagged with lighting directions. Remarkably, the generated shadows quickly communicate the underlying 3D structure of the sketched scene. Consequently, the shadows generated by our approach can be used directly or as an excellent starting point for artists. We demonstrate that the deep learning network we propose takes a handdrawn sketch, builds a 3D model in latent space, and renders the resulting shadows. The generated shadows respect the handdrawn lines and underlying 3D space and contain sophisticated and accurate details, such as selfshadowing effects. Moreover, the generated shadows contain artistic effects, such as rim lighting or halos appearing from back lighting, that would be achievable with traditional 3D rendering methods.

Simply generated noncrossing partitions ; We introduce and study the model of simply generated noncrossing partitions, which are, roughly speaking, chosen at random according to a sequence of weights. This framework encompasses the particular case of uniform noncrossing partitions with constraints on their block sizes. Our main tool is a bijection between noncrossing partitions and plane trees, which maps such simply generated noncrossing partitions into simply generated trees so that blocks of size k are in correspondence with vertices of outdegree k. This allows us to obtain limit theorems concerning the block structure of simply generated noncrossing partitions. We apply our results in free probability by giving a simple formula relating the maximum of the support of a compactly supported probability measure on the real line in term of its free cumulants.

MetricIndependent VolumeForms in Gravity and Cosmology ; Employing alternative spacetime volumeforms generallycovariant integration measure densities independent of the pertinent Riemannian spacetime metric have profound impact in general relativity. Although formally appearing as puregauge dynamical degrees of freedom they trigger a number of remarkable physically important phenomena such as i new mechanism of dynamical generation of cosmological constant; ii new type of quintessential inflation scenario in cosmology; iii nonsingular initial emergent universe phase of cosmological evolution preceding the inflationary phase; iv new mechanism of dynamical spontaneous breakdown of supersymmetry in supergravity; v gravitational electrovacuum bags. We study in some detail the properties, together with their canonical Hamiltonian formulation, of a class of generalized gravitymatter models built with two independent nonRiemannian volumeforms and discuss their implications in cosmology.

Scaleinvariant perturbations in ekpyrotic cosmologies without finetuning of initial conditions ; Ekpyrotic bouncing cosmologies have been proposed as alternatives to inflation. In these scenarios, the universe is smoothed and flattened during a period of slow contraction preceding the bounce while quantum fluctuations generate nearly scaleinvariant superhorizon perturbations that seed structure in the postbounce universe. An analysis by Tolley and Wesley 2007 showed that, for a wide range of ekpyrotic models, generating a scaleinvariant spectrum of adiabatic or entropic fluctuations is only possible if the cosmological background is unstable, in which case the scenario is highly sensitive to initial conditions. In this paper, we analyze an important counterexample a simple action that generates a Gaussian, scaleinvariant spectrum of entropic perturbations during ekpyrotic contraction without requiring finetuned initial conditions. Based on this example, we discuss some generalizations.

BuchdahlBondi limit in modified gravity Packing extra effective mass in relativistic compact stars ; We generalise the BuchdahlBondi limit for the case of static, spherically symmetric, relativistic compact stars immersed in Schwarzschild vacuum in fRtheory of gravity, subject to very generic regularity, thermodynamic stability and matching conditions. Similar to the case of general relativity, our result is model independent and remains true for any physically realistic equation of state of standard stellar matter. We show that an extramassive stable star can exist in these theories, with surface redshift larger than 2, which is forbidden in general relativity. This result gives a novel and interesting observational test for validity or otherwise of general relativity and also provides a possible solution to the dark matter problem.

Parameterized postNewtonian limit of Horndeski's gravity theory ; We discuss the parameterized postNewtonian PPN limit of Horndeski's theory of gravity, also known under the name generalized Ginflation or textG2inflation, which is the most general scalartensor theory of gravity with at most second order field equations in four dimensions. We derive conditions on the action for the validity of the postNewtonian limit. For the most general class of theories consistent with these conditions we calculate the PPN parameters gammar and betar, which in general depend on the interaction distance r between the gravitating mass and the test mass. For a more restricted class of theories, in which the scalar field is massless, we calculate the full set of PPN parameters. It turns out that in this restricted case all parameters are constants and that the only parameters potentially deviating from observations are gamma and beta. We finally apply our results to a number of example theories, including galileons and different models of Higgs inflation.

A Nonparametric Framework for Quantifying Generative Inference on Neuromorphic Systems ; Restricted Boltzmann Machines and Deep Belief Networks have been successfully used in probabilistic generative model applications such as image occlusion removal, pattern completion and motion synthesis. Generative inference in such algorithms can be performed very efficiently on hardware using a Markov Chain Monte Carlo procedure called Gibbs sampling, where stochastic samples are drawn from noisy integrate and fire neurons implemented on neuromorphic substrates. Currently, no satisfactory metrics exist for evaluating the generative performance of such algorithms implemented on highdimensional data for neuromorphic platforms. This paper demonstrates the application of nonparametric goodnessoffit testing to both quantify the generative performance as well as provide decisiondirected criteria for choosing the parameters of the neuromorphic Gibbs sampler and optimizing usage of hardware resources used during sampling.

Generalization of Quadratic Manifolds for Reduced Order Modeling of Nonlinear Structural Dynamics ; In this paper, a generalization of a quadratic manifold approach for the reduction of geometrically nonlinear structural dynamics problems is presented. This generalization is constructed by a linearization of the static force with respect to the generalized coordinates, resulting in a shift of the quadratic behavior from the force to the manifold. In this framework, static derivatives emerge as natural extensions to modal derivatives for displacement fields other than the vibration modes, such as the Krylov subspace vectors. Here the dynamic problem is projected onto the tangent space of the quadratic manifold, allowing for a much less number of generalized coordinates compared to linear basis methods. The potential of the quadratic manifold approach is investigated in a numerical study, where several variations of the approach are compared on different examples, indicating a clear pattern where the proposed approach is applicable.

DeepFace Face Generation using Deep Learning ; We use CNNs to build a system that both classifies images of faces based on a variety of different facial attributes and generates new faces given a set of desired facial characteristics. After introducing the problem and providing context in the first section, we discuss recent work related to image generation in Section 2. In Section 3, we describe the methods used to finetune our CNN and generate new images using a novel approach inspired by a Gaussian mixture model. In Section 4, we discuss our working dataset and describe our preprocessing steps and handling of facial attributes. Finally, in Sections 5, 6 and 7, we explain our experiments and results and conclude in the following section. Our classification system has 82 test accuracy. Furthermore, our generation pipeline successfully creates wellformed faces.

Learning Particle Physics by Example LocationAware Generative Adversarial Networks for Physics Synthesis ; We provide a bridge between generative modeling in the Machine Learning community and simulated physical processes in High Energy Particle Physics by applying a novel Generative Adversarial Network GAN architecture to the production of jet images 2D representations of energy depositions from particles interacting with a calorimeter. We propose a simple architecture, the LocationAware Generative Adversarial Network, that learns to produce realistic radiation patterns from simulated high energy particle collisions. The pixel intensities of GANgenerated images faithfully span over many orders of magnitude and exhibit the desired lowdimensional physical properties i.e., jet mass, nsubjettiness, etc.. We shed light on limitations, and provide a novel empirical validation of image quality and validity of GANproduced simulations of the natural world. This work provides a base for further explorations of GANs for use in faster simulation in High Energy Particle Physics.

On generalized semiMarkov quantum evolution ; We provide a large class of quantum evolution governed by the memory kernel master equation. This class defines quantum analog of so called semiMarkov classical stochastic evolution. In this Letter for the first time we provide a proper definition of quantum semiMarkov evolution and using the appropriate gauge freedom we propose a suitable generalization which contains majority of examples considered so far in the literature. The key concepts are quantum counterparts of classical waiting time distribution and survival probability quantum waiting time operator and quantum survival operator, respectively. In particular collision models and its generalizations considered recently are special examples of generalized semiMarkov evolution. This approach allows for an interesting generalization of trajectory description of the quantum dynamics in terms of POVM densities.

Black Hole Continuum Spectra as a Test of General Relativity Quadratic Gravity ; Observations of the continuum spectrum emitted by accretion disks around black holes allows us to infer their properties, including possibly whether black holes are described by the Kerr metric. Some modified gravity theories do not admit the Kerr metric as a solution, and thus, continuum spectrum observations could be used to constrain these theories. We here investigate whether current and next generation XRay observations of the black hole continuum spectrum can constrain such deviations from Einstein's theory, focusing on two wellmotivated modified quadratic gravity theories dynamical ChernSimons gravity and EinsteindilatonGaussBonnet gravity. We do so by determining whether the nonKerr deviations in the continuum spectrum introduced by these theories are larger than the observational error intrinsic to the observations. We find that dynamical ChernSimons gravity cannot be constrained better than current bounds with current or next generation continuum spectrum observations. EinsteindilatonGaussBonnet gravity, however, may be constrained better than current bounds with next generation telescopes, as long as the systematic error inherent in the accretion disk modeling is decreased below the predicted observational error.

A freely generated ring for N1 models in class Sk ; We study 4d N1 supersymmetric theories of class Sk, obtained from flux compactifications on a Riemann surface of 6d 1,0 conformal theories describing the low energy physics on a stack of M5 branes probing a Zk singularity. We conjecture that the protected spectrum of class Sk theories contains a freely generated ring, generalizing the Coulomb branch of the N2 theories. We derive this by examining a limit of the supersymmetric index of 4d N1 class Sk theories. The limit generalizes the Coulomb limit of N2 theories, which coincides with the case of k1 for a particular choice of flux. We conjecture a general simple formula for the index in the aforementioned limit.

RateDistortion Theory for General Sets and Measures ; This paper is concerned with a ratedistortion theory for sequences of i.i.d. random variables with general distribution supported on general sets including manifolds and fractal sets. Manifold structures are prevalent in data science, e.g., in compressed sensing, machine learning, image processing, and handwritten digit recognition. Fractal sets find application in image compression and in modeling of Ethernet traffic. We derive a lower bound on the singleletter ratedistortion function that applies to random variables X of general distribution and for continuous X reduces to the classical Shannon lower bound. Moreover, our lower bound is explicit up to a parameter obtained by solving a convex optimization problem in a nonnegative real variable. The only requirement for the bound to apply is the existence of a sigmafinite reference measure for X satisfying a certain subregularity condition. This condition is very general and prevents the reference measure from being highly concentrated on balls of small radii. To illustrate the wide applicability of our result, we evaluate the lower bound for a random variable distributed uniformly on a manifold, namely, the unit circle, and a random variable distributed uniformly on a selfsimilar set, namely, the middle third Cantor set.

The structure of generic anomalous dimensions and no theorem for massless propagators ; Extending an argument of Baikov2010hf for the case of 5loop massless propagators we prove a host of new exact modelindependent relations between contributions proportional to odd and even zetas in generic MSbar anomalous dimensions as well as in generic massless correlators. In particular, we find a new remarkable connection between coefficients in front of zeta3 and zeta4 in the 4loop and 5loop contributions to the QCD betafunction respectively. It leads to a natural explanation of a simple mechanics behind mysterious cancellations of the pidependent terms in onescale Renormalization Group RG invariant Euclidian quantities recently discovered in citeJamin2017mul. We give a proof of this nopi theorem for a general case of not necessarily schemeindependent onescale massless correlators. All pidependent terms in the bf sixloop coefficient of an anomalous dimension or a betafunction are shown to be explicitly expressible in terms of lower order coefficients for a general onecharge theory. For the case of a scalar On phi4 theory all our predictions for pidependent terms in 6loop anomalous dimensions are in full agreement with recent results of Batkovich2016jus,Schnetz2016fhy,Kompaniets2017yct.

Listen to Dance Musicdriven choreography generation using Autoregressive EncoderDecoder Network ; Automatic choreography generation is a challenging task because it often requires an understanding of two abstract concepts music and dance which are realized in the two different modalities, namely audio and video, respectively. In this paper, we propose a musicdriven choreography generation system using an autoregressive encoderdecoder network. To this end, we first collect a set of multimedia clips that include both music and corresponding dance motion. We then extract the joint coordinates of the dancer from video and the melspectrogram of music from audio, and train our network using musicchoreography pairs as input. Finally, a novel dance motion is generated at the inference time when only music is given as an input. We performed a user study for a qualitative evaluation of the proposed method, and the results show that the proposed model is able to generate musically meaningful and natural dance movements given an unheard song.

Codeswitching Sentence Generation by Generative Adversarial Networks and its Application to Data Augmentation ; Codeswitching is about dealing with alternative languages in speech or text. It is partially speakerdepend and domainrelated, so completely explaining the phenomenon by linguistic rules is challenging. Compared to most monolingual tasks, insufficient data is an issue for codeswitching. To mitigate the issue without expensive human annotation, we proposed an unsupervised method for codeswitching data augmentation. By utilizing a generative adversarial network, we can generate intrasentential codeswitching sentences from monolingual sentences. We applied proposed method on two corpora, and the result shows that the generated codeswitching sentences improve the performance of codeswitching language models.

SEIGAN Towards Compositional Image Generation by Simultaneously Learning to Segment, Enhance, and Inpaint ; We present a novel approach to image manipulation and understanding by simultaneously learning to segment object masks, paste objects to another background image, and remove them from original images. For this purpose, we develop a novel generative model for compositional image generation, SEIGAN SegmentEnhanceInpaint Generative Adversarial Network, which learns these three operations together in an adversarial architecture with additional cycle consistency losses. To train, SEIGAN needs only bounding box supervision and does not require pairing or ground truth masks. SEIGAN produces better generated images evaluated by human assessors than other approaches and produces highquality segmentation masks, improving over other adversarially trained approaches and getting closer to the results of fully supervised training.

Anisotropic stars as ultracompact objects in General Relativity ; Anisotropic stresses are ubiquitous in nature, but their modeling in General Relativity is poorly understood and frame dependent. We introduce the first study on the dynamical properties of anisotropic selfgravitating fluids in a covariant framework. Our description is particularly useful in the context of tests of the black hole paradigm, wherein ultracompact objects are used as black hole mimickers but otherwise lack a proper theoretical framework. We show that i anisotropic stars can be as compact and as massive as black holes, even for very small anisotropy parameters; ii the nonlinear dynamics of the 11 system is in good agreement with linearized calculations, and shows that configurations below the maximum mass are nonlinearly stable; iii strongly anisotropic stars have vanishing tidal Love numbers in the blackhole limit; iv their formation will usually be accompanied by gravitationalwave echoes at late times.

Segmentation Guided ImagetoImage Translation with Adversarial Networks ; Recently imagetoimage translation has received increasing attention, which aims to map images in one domain to another specific one. Existing methods mainly solve this task via a deep generative model, and focus on exploring the relationship between different domains. However, these methods neglect to utilize higherlevel and instancespecific information to guide the training process, leading to a great deal of unrealistic generated images of low quality. Existing methods also lack of spatial controllability during translation. To address these challenge, we propose a novel Segmentation Guided Generative Adversarial Networks SGGAN, which leverages semantic segmentation to further boost the generation performance and provide spatial mapping. In particular, a segmentor network is designed to impose semantic information on the generated images. Experimental results on multidomain face image translation task empirically demonstrate our ability of the spatial modification and our superiority in image quality over several stateoftheart methods.

Stochastic Backgrounds of Gravitational Waves from Cosmological Sources Techniques and Applications to Preheating ; Several mechanisms exist for generating a stochastic background of gravitational waves in the period following inflation. These mechanisms are generally classical in nature, with the gravitational waves being produced from inhomogeneities in the fields that populate the early universe and not quantum fluctuations. The resulting stochastic background could be accessible to next generation gravitational wave detectors. We develop a framework for computing such a background analytically and computationally. As an application of our framework, we consider the stochastic background of gravitational waves generated in a simple model of preheating.

Gravitational Radiation Generated by Cosmological Phase Transition Magnetic Fields ; We study gravitational waves generated by the cosmological magnetic fields induced via bubble collisions during the electroweak EW and QCD phase transitions. The magnetic field generation mechanisms considered here are based on the use of the fundamental EW minimal supersymmetric MSSM and QCD Lagrangians. The gravitational waves spectrum is computed using a magnetohydrodynamic MHD turbulence model. We find that gravitational wave spectrum amplitude generated by the EW phase transition peaks at frequency approximately 12 mHz, and is of the order of 10201021; thus this signal is possibly detectable by Laser Interferometer Space Antenna LISA. The gravitational waves generated during the QCD phase transition, however, are outside the LISA sensitivity bands.

Video2GIF Automatic Generation of Animated GIFs from Video ; We introduce the novel problem of automatically generating animated GIFs from video. GIFs are short looping video with no sound, and a perfect combination between image and video that really capture our attention. GIFs tell a story, express emotion, turn events into humorous moments, and are the new wave of photojournalism. We pose the question Can we automate the entirely manual and elaborate process of GIF creation by leveraging the plethora of user generated GIF content We propose a Robust Deep RankNet that, given a video, generates a ranked list of its segments according to their suitability as GIF. We train our model to learn what visual content is often selected for GIFs by using over 100K user generated GIFs and their corresponding video sources. We effectively deal with the noisy web data by proposing a novel adaptive Huber loss in the ranking formulation. We show that our approach is robust to outliers and picks up several patterns that are frequently present in popular animated GIFs. On our new largescale benchmark dataset, we show the advantage of our approach over several stateoftheart methods.

Generative Adversarial Text to Image Synthesis ; Automatic synthesis of realistic images from text would be interesting and useful, but current AI systems are still far from this goal. However, in recent years generic and powerful recurrent neural network architectures have been developed to learn discriminative text feature representations. Meanwhile, deep convolutional generative adversarial networks GANs have begun to generate highly compelling images of specific categories, such as faces, album covers, and room interiors. In this work, we develop a novel deep architecture and GAN formulation to effectively bridge these advances in text and image model ing, translating visual concepts from characters to pixels. We demonstrate the capability of our model to generate plausible images of birds and flowers from detailed text descriptions.

Opportunistic Advertisement Scheduling in Live Social Media A Multiple Stopping Time POMDP Approach ; Live online social broadcasting services like YouTube Live and Twitch have steadily gained popularity due to improved bandwidth, ease of generating content and the ability to earn revenue on the generated content. In contrast to traditional cable television, revenue in online services is generated solely through advertisements, and depends on the number of clicks generated. Channel owners aim to opportunistically schedule advertisements so as to generate maximum revenue. This paper considers the problem of optimal scheduling of advertisements in live online social media. The problem is formulated as a multiple stopping problem and is addressed in a partially observed Markov decision process POMDP framework. Structural results are provided on the optimal advertisement scheduling policy. By exploiting the structure of the optimal policy, best linear thresholds are computed using stochastic approximation. The proposed model and framework are validated on real datasets, and the following observations are made i The policy obtained by the multiple stopping problem can be used to detect changes in ground truth from online search data ii Numerical results show a significant improvement in the expected revenue by opportunistically scheduling the advertisements. The revenue can be improved by 2030 in comparison to currently employed periodic scheduling.

Investigation on Finsler geometry as a generalization to curved spacetime of Planckscaledeformed relativity in the de Sitter case ; In recent years, Planckscale modifications to particles' dispersion relation have been deeply studied for the possibility to formulate some phenomenology of Planckian effects in astrophysical and cosmological frameworks. There are some indications arXivgrqc0611024 that Finsler geometry can provide some generalization of Riemannian geometry which may allow to account for nontrivial Planckian structure of relativistic particles' configuration space. We investigate the possibility to formalize Planckscale deformations to relativistic models in curved spacetime, within the framework of Finsler geometry. We take into account the general strategy of analysis of modifications of dispersion relations in curved spacetimes proposed in arXiv1507.02056, generalizing to the de Sitter case the results obtained in arXiv1407.8143, for deformed relativistic particle kinematics in flat spacetime using Finsler formalism.

Generalized uncertainty principles, effective Newton constant and regular black holes ; In this paper, we explore the quantum spacetimes that are potentially connected with the generalized uncertainty principles. By analyzing the gravityinduced quantum interference pattern and the Gedanken for weighting photon, we find that the generalized uncertainty principles inspire the effective Newton constant as same as our previous proposal. A characteristic momentum associated with the tidal effect is suggested, which incorporates the quantum effect with the geometric nature of gravity. When the simplest generalized uncertainty principle is considered, the minimal model of the regular black holes is reproduced by the effective Newton constant. The black hole's tunneling probability, accurate to the second order correction, is carefully analyzed. We find that the tunneling probability is regularized by the size of the black hole remnant. Moreover, the black hole remnant is the final state of a tunneling process that the probability is minimized. A theory of modified gravity is suggested, by substituting the effective Newton constant into the HilbertEinstein action.

A generalized Lanczos method for systematic optimization of tensor network states ; We propose a generalized Lanczos method to generate the manybody basis states of quantum lattice models using tensornetwork states TNS. The groundstate wave function is represented as a linear superposition composed from a set of TNS generated by Lanczos iteration. This method improves significantly both the accuracy and the efficiency of the tensornetwork algorithm and allows the ground state to be determined accurately using TNS with very small virtual bond dimensions. This state contains significantly more entanglement than each individual TNS, reproducing correctly the logarithmic size dependence of the entanglement entropy in a critical system. The method can be generalized to nonHamiltonian systems and to the calculation of lowlying excited states, dynamical correlation functions, and other physical properties of strongly correlated systems.

Error Rates Analysis of MIMO SpaceTime Block Codes in Generalized Shadowed Fading Channels ; This paper introduces a new and unified bit error rates performance analysis of spacetime block codes STBC deployed in wireless systems with spatial diversity in generalized shadowed fading and noise scenarios. Specifically, we derive a simple and a very accurate approximate expressions for the average error rates of coherent modulation schemes in generalized etamu and kappamu shadowed fading channels with multiple input multiple output MIMO systems. The noise in the network is assumed to be modeled using the additive white generalized Gaussian noise AWGGN, which encompasses the classical Laplacian and the Gaussian noise environments as special cases. The derived results obviate the need to rederive the error rates for MIMO STBC systems under many multipath fading and noise conditions while avoiding any special functions with high computational complexity. Published results from the literature, as well as numerical evaluations, corroborate the accuracy of our derived generalized expressions.

A generative model for sparse, evolving digraphs ; Generating graphs that are similar to real ones is an open problem, while the similarity notion is quite elusive and hard to formalize. In this paper, we focus on sparse digraphs and propose SDG, an algorithm that aims at generating graphs similar to real ones. Since real graphs are evolving and this evolution is important to study in order to understand the underlying dynamical system, we tackle the problem of generating series of graphs. We propose SEDGE, an algorithm meant to generate series of graphs similar to a real series. SEDGE is an extension of SDG. We consider graphs that are representations of software programs and show experimentally that our approach outperforms other existing approaches. Experiments show the performance of both algorithms.

Detection of Generalized Synchronization using Echo State Networks ; Generalized synchronization between coupled dynamical systems is a phenomenon of relevance in applications that range from secure communications to physiological modelling. Here we test the capabilities of reservoir computing and, in particular, echo state networks for the detection of generalized synchronization. A nonlinear dynamical system consisting of two coupled Rossler chaotic attractors is used to generate temporal series consisting of timelocked generalized synchronized sequences interleaved by unsynchronized ones. Correctly tuned, echo state networks are able to efficiently discriminate between unsynchronized and synchronized sequences. Compared to other stateoftheart techniques of synchronization detection, the online capabilities of the proposed ESN based methodology make it a promising choice for realtime applications aiming to monitor dynamical synchronization changes in continuous signals.

An Unsupervised Deep Learning Approach for Scenario Forecasts ; In this paper, we propose a novel scenario forecasts approach which can be applied to a broad range of power system operations e.g., wind, solar, load over various forecasts horizons and prediction intervals. This approach is modelfree and datadriven, producing a set of scenarios that represent possible future behaviors based only on historical observations and point forecasts. It first applies a newlydeveloped unsupervised deep learning framework, the generative adversarial networks, to learn the intrinsic patterns in historical renewable generation data. Then by solving an optimization problem, we are able to quickly generate large number of realistic future scenarios. The proposed method has been applied to a wind power generation and forecasting dataset from national renewable energy laboratory. Simulation results indicate our method is able to generate scenarios that capture spatial and temporal correlations. Our code and simulation datasets are freely available online.

JamBot Music Theory Aware Chord Based Generation of Polyphonic Music with LSTMs ; We propose a novel approach for the generation of polyphonic music based on LSTMs. We generate music in two steps. First, a chord LSTM predicts a chord progression based on a chord embedding. A second LSTM then generates polyphonic music from the predicted chord progression. The generated music sounds pleasing and harmonic, with only few dissonant notes. It has clear longterm structure that is similar to what a musician would play during a jam session. We show that our approach is sensible from a music theory perspective by evaluating the learned chord embeddings. Surprisingly, our simple model managed to extract the circle of fifths, an important tool in music theory, from the dataset.

On the Automatic Generation of Medical Imaging Reports ; Medical imaging is widely used in clinical practice for diagnosis and treatment. Reportwriting can be errorprone for unexperienced physicians, and time consuming and tedious for experienced physicians. To address these issues, we study the automatic generation of medical imaging reports. This task presents several challenges. First, a complete report contains multiple heterogeneous forms of information, including findings and tags. Second, abnormal regions in medical images are difficult to identify. Third, the re ports are typically long, containing multiple sentences. To cope with these challenges, we 1 build a multitask learning framework which jointly performs the pre diction of tags and the generation of para graphs, 2 propose a coattention mechanism to localize regions containing abnormalities and generate narrations for them, 3 develop a hierarchical LSTM model to generate long paragraphs. We demonstrate the effectiveness of the proposed methods on two publicly available datasets.

Generative Adversarial Networks for Electronic Health Records A Framework for Exploring and Evaluating Methods for Predicting DrugInduced Laboratory Test Trajectories ; Generative Adversarial Networks GANs represent a promising class of generative networks that combine neural networks with game theory. From generating realistic images and videos to assisting musical creation, GANs are transforming many fields of arts and sciences. However, their application to healthcare has not been fully realized, more specifically in generating electronic health records EHR data. In this paper, we propose a framework for exploring the value of GANs in the context of continuous laboratory time series data. We devise an unsupervised evaluation method that measures the predictive power of synthetic laboratory test time series. Further, we show that when it comes to predicting the impact of drug exposure on laboratory test data, incorporating representation learning of the training cohorts prior to training GAN models is beneficial.

Visual to Sound Generating Natural Sound for Videos in the Wild ; As two of the five traditional human senses sight, hearing, taste, smell, and touch, vision and sound are basic sources through which humans understand the world. Often correlated during natural events, these two modalities combine to jointly affect human perception. In this paper, we pose the task of generating sound given visual input. Such capabilities could help enable applications in virtual reality generating sound for virtual scenes automatically or provide additional accessibility to images or videos for people with visual impairments. As a first step in this direction, we apply learningbased methods to generate raw waveform samples given input video frames. We evaluate our models on a dataset of videos containing a variety of sounds such as ambient sounds and sounds from peopleanimals. Our experiments show that the generated sounds are fairly realistic and have good temporal synchronization with the visual inputs.

Generalization of Deep Neural Networks for Chest Pathology Classification in XRays Using Generative Adversarial Networks ; Medical datasets are often highly imbalanced with overrepresentation of common medical problems and a paucity of data from rare conditions. We propose simulation of pathology in images to overcome the above limitations. Using chest Xrays as a model medical image, we implement a generative adversarial network GAN to create artificial images based upon a modest sized labeled dataset. We employ a combination of real and artificial images to train a deep convolutional neural network DCNN to detect pathology across five classes of chest Xrays. Furthermore, we demonstrate that augmenting the original imbalanced dataset with GAN generated images improves performance of chest pathology classification using the proposed DCNN in comparison to the same DCNN trained with the original dataset alone. This improved performance is largely attributed to balancing of the dataset using GAN generated images, where image classes that are lacking in example images are preferentially augmented.

An Improved Feedback Coding Scheme for the Wiretap Channel ; The model of wiretap channel WTC is important as it constitutes the essence of physical layer security PLS. Wiretap channel with noiseless feedback WTCNF is especially interesting as it shows what can be done when a private feedback is available. The already existing secret key based feedback coding scheme focuses on generating key from the feedback and using this key to protect part of the transmitted message. It has been shown that this secret key based feedback coding scheme is only optimal for the degraded WTCNF, and finding an optimal feedback scheme for the general WTCNF motivates us to exploit other uses of the feedback. In this paper, a new feedback coding scheme for the general WTCNF is proposed, where the feedback is not only used to generate key, but also used to generate help information which helps the legitimate parties to improve the communication between them. We show that the proposed new feedback scheme performs better than the already existing one, and a binary example is given to further explain the results of this paper.

Transition from wakefield generation to soliton formation ; It is well known that when a short laser pulse propagates in an underdense plasma, it induces longitudinal plasma oscillations at the plasma frequency after the pulse, typically referred to as the 'wakefield'. However, for plasma densities approaching the critical density wakefield generation is suppressed, and instead the EMpulse undergoes nonlinear selfmodulation. In this article we have studied the transition from the wakefield generation to formation of quasisolitons as the plasma density is increased. For this purpose we have applied a one dimensional 1D relativistic cold fluid model, which has also been compared with particleincell simulations. A key result is that the energy loss of the EMpulse due to wakefield generation has its maximum for a plasma density of the order 10 percent of the critical density, but that wakefield generation is sharply suppressed when the density is increased further.

Coalgebraic Behavioral Metrics ; We study different behavioral metrics, such as those arising from both branching and lineartime semantics, in a coalgebraic setting. Given a coalgebra alphacolon X to HX for a functor H colon mathrmSetto mathrmSet, we define a framework for deriving pseudometrics on X which measure the behavioral distance of states. A crucial step is the lifting of the functor H on mathrmSet to a functor overlineH on the category mathrmPMet of pseudometric spaces. We present two different approaches which can be viewed as generalizations of the Kantorovich and Wasserstein pseudometrics for probability measures. We show that the pseudometrics provided by the two approaches coincide on several natural examples, but in general they differ. If H has a final coalgebra, every lifting overlineH yields in a canonical way a behavioral distance which is usually branchingtime, i.e., it generalizes bisimilarity. In order to model lineartime metrics generalizing trace equivalences, we show sufficient conditions for lifting distributive laws and monads. These results enable us to employ the generalized powerset construction.

Recognizing Generalized Transmission Graphs of Line Segments and Circular Sectors ; Suppose we have an arrangement mathcalA of n geometric objects x1, dots, xn subseteq mathbbR2 in the plane, with a distinguished point pi in each object xi. The generalized transmission graph of mathcalA has vertex set x1, dots, xn and a directed edge xixj if and only if pj in xi. Generalized transmission graphs provide a generalized model of the connectivity in networks of directional antennas. The complexity class exists mathbbR contains all problems that can be reduced in polynomial time to an existential sentence of the form exists x1, dots, xn phix1,dots, xn, where x1,dots, xn range over mathbbR and phi is a propositional formula with signature , , cdot, 0, 1. The class exists mathbbR aims to capture the complexity of the existential theory of the reals. It lies between mathbfNP and mathbfPSPACE. Many geometric decision problems, such as recognition of disk graphs and of intersection graphs of lines, are complete for exists mathbbR. Continuing this line of research, we show that the recognition problem of generalized transmission graphs of line segments and of circular sectors is hard for exists mathbbR. As far as we know, this constitutes the first such result for a class of directed graphs.

A point mass and continuous collapse to a point mass in general relativity ; An original way of presentation of the Schwarzschild black hole in the form of a pointlike mass with making the use of the Dirac deltafunction, including a description of a continuous collapse to such a point mass, is given. A maximally generalized description restricted by physically reasonable requirements is developed. A socalled fieldtheoretical formulation of general relativity, being equivalent to the standard geometrical presentation of general relativity, is used. All of the dynamical fields, including the gravitational field, are considered as propagating in a background curved or flat spacetime. Namely these properties allow us to present a noncontradictive picture of the point mass description. The results can be useful for studying the structure of the black hole true singularities and could be developed for practical calculations in models with black holes.

Ternary generalization of Pauli's principle and the Z6graded algebras ; We show how the discrete symmetries Z2 and Z3 combined with the superposition principle result in the SL2, bf Csymmetry of quantum states. The role of Pauli's exclusion principle in the derivation of the SL2, C symmetry is put forward as the source of the macroscopically observed Lorentz symmetry, then it is generalized for the case of the Z3 grading replacing the usual Z2 grading, leading to ternary commutation relations. We discuss the cubic and ternary generalizations of Grassmann algebra. Invariant cubic forms are introduced, and their symmetry group is shown to be the SL2,C group The wave equation generalizing the Dirac operator to the Z3graded case is constructed. Its diagonalization leads to a sixthorder equation. The solutions cannot propagate because their exponents always contain nonoscillating real damping factor. We show how certain cubic products can propagate nevertheless. The model suggests the origin of the color SU3 symmetry.

Laserinduced plasma generation of terahertz radiation using three incommensurate wavelengths ; We present the generation of THz radiation by focusing ultrafast laser pulses with three incommensurate wavelengths to form a plasma. The three colors include 800 nm and the variable IR signal and idler outputs from an optical parametric amplifier. Stable THz is generated when all three colors are present, with a peaktopeak field strength of 200 kVcm and a relatively broad, smooth spectrum extending out to 6 THz, without any strong dependence on the selection of signal and idler IR wavelengths in the range from 13002000 nm. We confirm that 3 colors are indeed needed, and comment on the polarization characteristics of the generated THz, some of which are challenging to explain with plasma current models that have had success in describing twocolor plasma THz generation.

Generative Sensing Transforming Unreliable Sensor Data for Reliable Recognition ; This paper introduces a deep learning enabled generative sensing framework which integrates lowend sensors with computational intelligence to attain a high recognition accuracy on par with that attained with highend sensors. The proposed generative sensing framework aims at transforming lowend, lowquality sensor data into higher quality sensor data in terms of achieved classification accuracy. The lowend data can be transformed into higher quality data of the same modality or into data of another modality. Different from existing methods for image generation, the proposed framework is based on discriminative models and targets to maximize the recognition accuracy rather than a similarity measure. This is achieved through the introduction of selective feature regeneration in a deep neural network DNN. The proposed generative sensing will essentially transform lowquality sensor data into highquality information for robust perception. Results are presented to illustrate the performance of the proposed framework.

Anisotropic generalization of wellknown solutions describing relativistic selfgravitating fluid systems An algorithm ; We present an algorithm to generalize a plethora of wellknown solutions to Einstein field equations describing spherically symmetric relativistic fluid spheres by relaxing the pressure isotropy condition on the system. By suitably fixing the model parameters in our formulation, we generate closedform solutions which may be treated as anisotropic generalization of a large class of solutions describing isotropic fluid spheres. From the resultant solutions, a particular solution is taken up to show its physical acceptability. Making use of the current estimate of mass and radius of a known pulsar, the effects of anisotropic stress on the gross physical behaviour of a relativistic compact star is also highlighted.

Midinfrared frequency comb generation via cascaded quadratic nonlinearities in quasiphasematched waveguides ; We experimentally demonstrate a simple configuration for midinfrared MIR frequency comb generation in quasiphasematched lithium niobate waveguides using the cascadedchi2 nonlinearity. With nanojoulescale pulses from an Erfiber laser, we observe octavespanning supercontinuum in the nearinfrared with dispersivewave generation in the 2.53 textmum region and intrapulse differencefrequency generation in the 45 textmum region. By engineering the quasiphasematched grating profiles, tunable, narrowband MIR and broadband MIR spectra are both observed in this geometry. Finally, we perform numerical modeling using a nonlinear envelope equation, which shows good quantitative agreement with the experimentand can be used to inform waveguide designs to tailor the MIR frequency combs. Our results identify a path to a simple singlebranch approach to midinfrared frequency comb generation in a compact platform using commercial Erfiber technology.

Conformal blocks from Wilson lines with loop corrections ; We compute the conformal blocks of the Virasoro minimal model or its WN extension with large central charge from Wilson line networks in a ChernSimons theory including loop corrections. In our previous work, we offered a prescription to regularize divergences from loops attached to Wilson lines. In this paper, we generalize our method with the prescription by dealing with more general operators for N3 and apply it to the identity W3 block. We further compute general lightlight blocks and heavylight correlators for N2 with the Wilson line method and compare the results with known ones obtained using a different prescription. We briefly discuss general W3 blocks.

Interactive Generative Adversarial Networks for Facial Expression Generation in Dyadic Interactions ; A social interaction is a social exchange between two or more individuals,where individuals modify and adjust their behaviors in response to their interaction partners. Our social interactions are one of most fundamental aspects of our lives and can profoundly affect our mood, both positively and negatively. With growing interest in virtual reality and avatarmediated interactions,it is desirable to make these interactions natural and human like to promote positive effect in the interactions and applications such as intelligent tutoring systems, automated interview systems and elearning. In this paper, we propose a method to generate facial behaviors for an agent. These behaviors include facial expressions and head pose and they are generated considering the users affective state. Our models learn semantically meaningful representations of the face and generate appropriate and temporally smooth facial behaviors in dyadic interactions.

BAGAN Data Augmentation with Balancing GAN ; Image classification datasets are often imbalanced, characteristic that negatively affects the accuracy of deeplearning classifiers. In this work we propose balancing GAN BAGAN as an augmentation tool to restore balance in imbalanced datasets. This is challenging because the few minorityclass images may not be enough to train a GAN. We overcome this issue by including during the adversarial training all available images of majority and minority classes. The generative model learns useful features from majority classes and uses these to generate images for minority classes. We apply class conditioning in the latent space to drive the generation process towards a target class. The generator in the GAN is initialized with the encoder module of an autoencoder that enables us to learn an accurate classconditioning in the latent space. We compare the proposed methodology with stateoftheart GANs and demonstrate that BAGAN generates images of superior quality when trained with an imbalanced dataset.

Experimental Implementation of Generalized Transitionless Quantum Driving ; It is known that high intensity fields are usually required to implement shortcuts to adiabaticity via Transitionless Quantum Driving TQD. Here, we show that this requirement can be relaxed by exploiting the gauge freedom of generalized TQD, which is expressed in terms of an arbitrary phase when mimicking the adiabatic evolution. We experimentally investigate the performance of generalized TQD in comparison with both traditional TQD and adiabatic dynamics. By using a 171Yb trapped ion hyperfine qubit, we implement a LandauZener adiabatic Hamiltonian and its traditional and generalized TQD counterparts. We show that the generalized theory provides optimally implementable Hamiltonians for TQD, with no additional fields required. In addition, the energetically optimal TQD Hamiltonian for the LandauZener model is investigated under dephasing. Remarkably, even using less intense fields, optimal TQD exhibits fidelities that are more robust against a decohering environment, with performance superior than that provided by the adiabatic dynamics.

Pioneer Networks Progressively Growing Generative Autoencoder ; We introduce a novel generative autoencoder network model that learns to encode and reconstruct images with high quality and resolution, and supports smooth random sampling from the latent space of the encoder. Generative adversarial networks GANs are known for their ability to simulate random highquality images, but they cannot reconstruct existing images. Previous works have attempted to extend GANs to support such inference but, so far, have not delivered satisfactory highquality results. Instead, we propose the Progressively Growing Generative Autoencoder PIONEER network which achieves highquality reconstruction with 128times128 images without requiring a GAN discriminator. We merge recent techniques for progressively building up the parts of the network with the recently introduced adversarial encodergenerator network. The ability to reconstruct input images is crucial in many realworld applications, and allows for precise intelligent manipulation of existing images. We show promising results in image synthesis and inference, with stateoftheart results in CelebA inference tasks.

HighResolution Mammogram Synthesis using Progressive Generative Adversarial Networks ; The ability to generate synthetic medical images is useful for data augmentation, domain transfer, and outofdistribution detection. However, generating realistic, highresolution medical images is challenging, particularly for Full Field Digital Mammograms FFDM, due to the textural heterogeneity, fine structural details and specific tissue properties. In this paper, we explore the use of progressively trained generative adversarial networks GANs to synthesize mammograms, overcoming the underlying instabilities when training such adversarial models. This work is the first to show that generation of realistic synthetic medical images is feasible at up to 1280x1024 pixels, the highest resolution achieved for medical image synthesis, enabling visualizations within standard mammographic hanging protocols. We hope this work can serve as a useful guide and facilitate further research on GANs in the medical imaging domain.

Doubly Stochastic Adversarial Autoencoder ; Any autoencoder network can be turned into a generative model by imposing an arbitrary prior distribution on its hidden code vector. Variational Autoencoder VAE 2 uses a KL divergence penalty to impose the prior, whereas Adversarial Autoencoder AAE 1 uses it generative adversarial networks GAN 3. GAN trades the complexities of it sampling algorithms with the complexities of it searching Nash equilibrium in minimax games. Such minimax architectures get trained with the help of data examples and gradients flowing through a generator and an adversary. A straightforward modification of AAE is to replace the adversary with the maximum mean discrepancy MMD test 45. This replacement leads to a new type of probabilistic autoencoder, which is also discussed in our paper. We propose a novel probabilistic autoencoder in which the adversary of AAE is replaced with a space of it stochastic functions. This replacement introduces a new source of randomness, which can be considered as a continuous control for encouraging it explorations. This prevents the adversary from fitting too closely to the generator and therefore leads to a more diverse set of generated samples.

Tensor chain and constraints in tensor networks ; This paper accompanies with our recent work on quantum error correction QEC and entanglement spectrum ES in tensor networks arXiv1806.05007. We propose a general framework for planar tensor network state with tensor constraints as a model for AdS3CFT2 correspondence, which could be viewed as a generalization of hyperinvariant tensor networks recently proposed by Evenbly. We elaborate our proposal on tensor chains in a tensor network by tiling H2 space and provide a diagrammatical description for general multitensor constraints in terms of tensor chains, which forms a generalized greedy algorithm. The behavior of tensor chains under the action of greedy algorithm is investigated in detail. In particular, for a given set of tensor constraints, a critically protected CP tensor chain can be figured out and evaluated by its average reduced interior angle. We classify tensor networks according to their ability of QEC and the flatness of ES. The corresponding geometric description of critical protection over the hyperbolic space is also given.

Residualbased iterations for the generalized Lyapunov equation ; This paper treats iterative solution methods to the generalized Lyapunov equation. Specifically it expands the existing theoretical justification for the alternating linear scheme ALS from the stable Lyapunov equation to the stable generalized Lyapunov equation. Moreover, connections between the energynorm minimization in ALS and the theory to H2optimality of an associated bilinear control system are established. It is also shown that a certain ALSbased iteration can be seen as iteratively constructing rank1 model reduction subspaces for bilinear control systems associated with the residual. Similar to the ALSbased iteration, the fixedpoint iteration can also be seen as a residualbased method minimizing an upper bound of the associated energy norm. Lastly a residualbased generalized rationalKrylovtype subspace is proposed for the generalized Lyapunov equation.

REFUGE CHALLENGE 2018Task 2Deep Optic Disc and Cup Segmentation in Fundus Images Using UNet and Multiscale Feature Matching Networks ; In this paper, an optic disc and cup segmentation method is proposed using UNet followed by a multiscale feature matching network. The proposed method targets task 2 of the REFUGE challenge 2018. In order to solve the segmentation problem of task 2, we firstly crop the input image using single shot multibox detector SSD. The cropped image is then passed to an encoderdecoder network with skip connections also known as generator. Afterwards, both the ground truth and generated images are fed to a convolution neural network CNN to extract their multilevel features. A dice loss function is then used to match the features of the two images by minimizing the error at each layer. The aggregation of error from each layer is backpropagated through the generator network to enforce it to generate a segmented image closer to the ground truth. The CNN network improves the performance of the generator network without increasing the complexity of the model.

Dynamically generated hierarchies in games of competition ; Spatial manyspecies predatorprey systems have been shown to yield very rich spacetime patterns. This observation begs the question whether there exist universal mechanisms for generating this type of emerging complex patterns in nonequilibrium systems. In this work we investigate the possibility of dynamically generated hierarchies in predatorprey systems. We analyze a ninespecies model with competing interactions and show that the studied situation results in the spontaneous formation of spirals within spirals. The parameter dependence of these intriguing nested spirals is elucidated. This is achieved through the numerical investigation of various quantities correlation lengths, densities of empty sites, Fourier analysis of species densities, interface fluctuations that allows to gain a rather complete understanding of the spatial arrangements and the temporal evolution of the system. A possible generalization of the interaction scheme yielding dynamically generated hierarchies is discussed. As cyclic interactions occur spontaneously in systems with competing strategies, the mechanism discussed in this work should contribute to our understanding of various social and biological systems.

A Look at the Effect of Sample Design on Generalization through the Lens of Spectral Analysis ; This paper provides a general framework to study the effect of sampling properties of training data on the generalization error of the learned machine learning ML models. Specifically, we propose a new spectral analysis of the generalization error, expressed in terms of the power spectra of the sampling pattern and the function involved. The framework is build in the Euclidean space using Fourier analysis and establishes a connection between some high dimensional geometric objects and optimal spectral form of different stateoftheart sampling patterns. Subsequently, we estimate the expected error bounds and convergence rate of different stateoftheart sampling patterns, as the number of samples and dimensions increase. We make several observations about generalization error which are valid irrespective of the approximation scheme or learning architecture and training or optimization algorithms. Our result also sheds light on ways to formulate design principles for constructing optimal sampling methods for particular problems.

Global optimization of dielectric metasurfaces using a physicsdriven neural network ; We present a global optimizer, based on a conditional generative neural network, which can output ensembles of highly efficient topologyoptimized metasurfaces operating across a range of parameters. A key feature of the network is that it initially generates a distribution of devices that broadly samples the design space, and then shifts and refines this distribution towards favorable design space regions over the course of optimization. Training is performed by calculating the forward and adjoint electromagnetic simulations of outputted devices and using the subsequent efficiency gradients for backpropagation. With metagratings operating across a range of wavelengths and angles as a model system, we show that devices produced from the trained generative network have efficiencies comparable to or better than the best devices produced by adjointbased topology optimization, while requiring less computational cost. Our reframing of adjointbased optimization to the training of a generative neural network applies generally to physical systems that can utilize gradients to improve performance.

The Effect of the Uncertainty of Load and Renewable Generation on the Dynamic Voltage Stability Margin ; In this paper, the impact of stochastic load and renewable generation uncertainty on the dynamic voltage stability margin is studied. Stochastic trajectories describing the uncertainty of load, wind and solar generation have been incorporated in the power system model as a set of Stochastic DifferentialAlgebraic Equations SDAEs. A systematic study of Monte Carlo dynamic simulations on the IEEE 39Bus system has been conducted to compute the stochastic load margin with all dynamic components active. Numerical results show that the uncertainty of both demand and generation may lead to a decrease on the size of the dynamic voltage stability margin, yet the variability of renewable generators may play a more significant role. Given that the integration of renewable energy will continue growing, it is of paramount importance to apply stochastic and dynamic approaches in the voltage stability study.

On improving deep learning generalization with adaptive sparse connectivity ; Large neural networks are very successful in various tasks. However, with limited data, the generalization capabilities of deep neural networks are also very limited. In this paper, we empirically start showing that intrinsically sparse neural networks with adaptive sparse connectivity, which by design have a strict parameter budget during the training phase, have better generalization capabilities than their fullyconnected counterparts. Besides this, we propose a new technique to train these sparse models by combining the Sparse Evolutionary Training SET procedure with neurons pruning. Operated on MultiLayer Perceptron MLP and tested on 15 datasets, our proposed technique zeros out around 50 of the hidden neurons during training, while having a linear number of parameters to optimize with respect to the number of neurons. The results show a competitive classification and generalization performance.

Bounded Displacement NonEquivalence In Substitution Tilings ; In the study of aperiodic order and mathematical models of quasicrystals, questions regarding equivalence relations on Delone sets naturally arise. This work is dedicated to the bounded displacement BD equivalence relation, and especially to results concerning instances of nonequivalence. We present a general condition for two Delone sets to be BD nonequivalent, and apply our result to Delone sets associated with tilings of Euclidean space. First we consider substitution tilings, and exhibit a substitution matrix associated with two distinct substitution rules. The first rule generates only periodic tilings, while the second generates tilings for which any associated Delone set is nonequivalent to any lattice in space. As an extension of this result, we introduce arbitrarily many distinct substitution rules associated with a single matrix, with the property that Delone sets generated by distinct rules are nonequivalent. We then turn to the study of mixed substitution tilings, and present a mixed substitution system that generates representatives of continuously many distinct BD equivalence classes.

Secure Network Coding in the Setting in Which a NonSource Node May Generate Random Keys ; It is common in the study of secure multicast network coding in the presence of an eavesdropper that has access to z network links, to assume that the source node is the only node that generates random keys. In this setting, the secure multicast rate is well understood. Computing the secure multicast rate, or even the secure unicast rate, in the more general setting in which all network nodes may generate independent random keys is known to be as difficult as computing the nonsecure capacity of multipleunicast network coding instances a well known open problem. This work treats an intermediate model of secure unicast in which only one node can generate random keys, however that node need not be the source node. The secure communication rate for this setting is characterized again with an eavesdropper that has access to z network links.

Melody Generation using an Interactive Evolutionary Algorithm ; Music generation with the aid of computers has been recently grabbed the attention of many scientists in the area of artificial intelligence. Deep learning techniques have evolved sequence production methods for this purpose. Yet, a challenging problem is how to evaluate generated music by a machine. In this paper, a methodology has been developed based upon an interactive evolutionary optimization method, with which the scoring of the generated melodies is primarily performed by human expertise, during the training. This music quality scoring is modeled using a BiLSTM recurrent neural network. Moreover, the innovative generated melody through a Genetic algorithm will then be evaluated using this BiLSTM network. The results of this mechanism clearly show that the proposed method is able to create pleasurable melodies with desired styles and pieces. This method is also quite fast, compared to the stateoftheart dataoriented evolutionary systems.

Domain Generalization via Multidomain Discriminant Analysis ; Domain generalization DG aims to incorporate knowledge from multiple source domains into a single model that could generalize well on unseen target domains. This problem is ubiquitous in practice since the distributions of the target data may rarely be identical to those of the source data. In this paper, we propose Multidomain Discriminant Analysis MDA to address DG of classification tasks in general situations. MDA learns a domaininvariant feature transformation that aims to achieve appealing properties, including a minimal divergence among domains within each class, a maximal separability among classes, and overall maximal compactness of all classes. Furthermore, we provide the bounds on excess risk and generalization error by learning theory analysis. Comprehensive experiments on synthetic and real benchmark datasets demonstrate the effectiveness of MDA.

Killer Technologies the destructive creation in the technical change ; Killer technology is a radical innovation, based on new products andor processes, that with high technical andor economic performance destroys the usage value of established techniques previously sold and used. Killer technology is a new concept in economics of innovation that may be useful for bringing a new perspective to explain and generalize the behavior and characteristics of innovations that generate a destructive creation for sustaining technical change. To explore the behavior of killer technologies, a simple model is proposed to analyze and predict how killer technologies destroy and substitute established technologies. Empirical evidence of this theoretical framework is based on historical data on the evolution of some example technologies. Theoretical framework and empirical evidence hint at general properties of the behavior of killer technologies to explain corporate, industrial, economic and social change and to support best practices for technology management of firms and innovation policy of nations. Overall, then, the proposed theoretical framework can lay a foundation for the development of more sophisticated concepts to explain the behavior of vital technologies that generate technological and industrial change in society.

Monotonic and NonMonotonic Solution Concepts for Generalized Circuits ; Generalized circuits are an important tool in the study of the computational complexity of equilibrium approximation problems. However, in this paper, we reveal that they have a conceptual flaw, namely that the solution concept is not monotonic. By this we mean that if varepsilon varepsilon', then an varepsilonapproximate solution for a certain generalized circuit is not necessarily also an varepsilon'approximate solution. The reason for this nonmonotonicity is the way Boolean operations are modeled. We illustrate that nonmonotonicity creates subtle technical issues in prior work that require intricate additional arguments to circumvent. To eliminate this problem, we show that the Boolean gates are a redundant feature one can simulate stronger, monotonic versions of the Boolean gates using the other gate types. Arguing at the level of these stronger Boolean gates eliminates all of the aforementioned issues in a natural way. We hope that our results will enable new studies of subclasses of generalized circuits and enabler simpler and more natural reductions from generalized circuits to other equilibrium search problems.

Unlimited Resolution Image Generation with R2D2GANs ; In this paper we present a novel simulation technique for generating high quality images of any predefined resolution. This method can be used to synthesize sonar scans of size equivalent to those collected during a fulllength mission, with across track resolutions of any chosen magnitude. In essence, our model extends Generative Adversarial Networks GANs based architecture into a conditional recursive setting, that facilitates the continuity of the generated images. The data produced is continuous, realisticallylooking, and can also be generated at least two times faster than the real speed of acquisition for the sonars with higher resolutions, such as EdgeTech. The seabed topography can be fully controlled by the user. The visual assessment tests demonstrate that humans cannot distinguish the simulated images from real. Moreover, experimental results suggest that in the absence of real data the autonomous recognition systems can benefit greatly from training with the synthetic data, produced by the R2D2GANs.

Adversarial Vertex Mixup Toward Better Adversarially Robust Generalization ; Adversarial examples cause neural networks to produce incorrect outputs with high confidence. Although adversarial training is one of the most effective forms of defense against adversarial examples, unfortunately, a large gap exists between test accuracy and training accuracy in adversarial training. In this paper, we identify Adversarial Feature Overfitting AFO, which may cause poor adversarially robust generalization, and we show that adversarial training can overshoot the optimal point in terms of robust generalization, leading to AFO in our simple Gaussian model. Considering these theoretical results, we present soft labeling as a solution to the AFO problem. Furthermore, we propose Adversarial Vertex mixup AVmixup, a softlabeled data augmentation approach for improving adversarially robust generalization. We complement our theoretical analysis with experiments on CIFAR10, CIFAR100, SVHN, and Tiny ImageNet, and show that AVmixup significantly improves the robust generalization performance and that it reduces the tradeoff between standard accuracy and adversarial robustness.

Generation of highorder harmonics with tunable photon energy and spectral width using double pulses ; This work theoretically investigates highorder harmonic generation in rare gas atoms driven by two temporally delayed ultrashort laser pulses. Apart from their temporal delay, the two pulses are identical. Using a singleatom model of the lasermatter interaction it is shown that the photon energy of the generated harmonics is controllable within the range of one eV a bandwidth comparable to the photon energy of the fundamental field by varying the time delay between the generating laser pulses. It is also demonstrated that highorder harmonics generated by double pulses have advantageous characteristics, which mimick certain properties of an extreme ultraviolet XUV monochromator. With the proposed method, a simpler setup at a much lower cost and comparatively higher spectral yield can be implemented in contrast to other approaches.

A General Approach for Using Deep Neural Network for Digital Watermarking ; Technologies of the Internet of Things IoT facilitate digital contents such as images being acquired in a massive way. However, consideration from the privacy or legislation perspective still demands the need for intellectual content protection. In this paper, we propose a general deep neural network DNN based watermarking method to fulfill this goal. Instead of training a neural network for protecting a specific image, we train on an image set and use the trained model to protect a distinct test image set in a bulk manner. Respective evaluations both from the subjective and objective aspects confirm the supremacy and practicability of our proposed method. To demonstrate the robustness of this general neural watermarking mechanism, commonly used manipulations are applied to the watermarked image to examine the corresponding extracted watermark, which still retains sufficient recognizable traits. To the best of our knowledge, we are the first to propose a general way to perform watermarking using DNN. Considering its performance and economy, it is concluded that subsequent studies that generalize our work on utilizing DNN for intellectual content protection is a promising research trend.

Lesion Conditional Image Generation for Improved Segmentation of Intracranial Hemorrhage from CT Images ; Data augmentation can effectively resolve a scarcity of images when training machinelearning algorithms. It can make them more robust to unseen images. We present a lesion conditional Generative Adversarial Network LcGAN to generate synthetic Computed Tomography CT images for data augmentation. A lesion conditional image segmented mask is an input to both the generator and the discriminator of the LcGAN during training. The trained model generates contextual CT images based on input masks. We quantify the quality of the images by using a fully convolutional network FCN score and blurriness. We also train another classification network to select better synthetic images. These synthetic CT images are then augmented to our hemorrhagic lesion segmentation network. By applying this augmentation method on 2.5, 10 and 25 of original data, segmentation improved by 12.8, 6 and 1.6 respectively.

Take the Scenic Route Improving Generalization in VisionandLanguage Navigation ; In the VisionandLanguage Navigation VLN task, an agent with egocentric vision navigates to a destination given natural language instructions. The act of manually annotating these instructions is timely and expensive, such that many existing approaches automatically generate additional samples to improve agent performance. However, these approaches still have difficulty generalizing their performance to new environments. In this work, we investigate the popular RoomtoRoom R2R VLN benchmark and discover that what is important is not only the amount of data you synthesize, but also how you do it. We find that shortest path sampling, which is used by both the R2R benchmark and existing augmentation methods, encode biases in the action space of the agent which we dub as action priors. We then show that these action priors offer one explanation toward the poor generalization of existing works. To mitigate such priors, we propose a path sampling method based on random walks to augment the data. By training with this augmentation strategy, our agent is able to generalize better to unknown environments compared to the baseline, significantly improving model performance in the process.

Quantum State Tomography with Conditional Generative Adversarial Networks ; Quantum state tomography QST is a challenging task in intermediatescale quantum devices. Here, we apply conditional generative adversarial networks CGANs to QST. In the CGAN framework, two duelling neural networks, a generator and a discriminator, learn multimodal models from data. We augment a CGAN with custom neuralnetwork layers that enable conversion of output from any standard neural network into a physical density matrix. To reconstruct the density matrix, the generator and discriminator networks train each other on data using standard gradientbased methods. We demonstrate that our QSTCGAN reconstructs optical quantum states with high fidelity orders of magnitude faster, and from less data, than a standard maximumlikelihood method. We also show that the QSTCGAN can reconstruct a quantum state in a single evaluation of the generator network if it has been pretrained on similar quantum states.

POP909 A Popsong Dataset for Music Arrangement Generation ; Music arrangement generation is a subtask of automatic music generation, which involves reconstructing and reconceptualizing a piece with new compositional techniques. Such a generation process inevitably requires reference from the original melody, chord progression, or other structural information. Despite some promising models for arrangement, they lack more refined data to achieve better evaluations and more practical results. In this paper, we propose POP909, a dataset which contains multiple versions of the piano arrangements of 909 popular songs created by professional musicians. The main body of the dataset contains the vocal melody, the lead instrument melody, and the piano accompaniment for each song in MIDI format, which are aligned to the original audio files. Furthermore, we provide the annotations of tempo, beat, key, and chords, where the tempo curves are handlabeled and others are done by MIR algorithms. Finally, we conduct several baseline experiments with this dataset using standard deep music generation algorithms.

AutoSimulate Quickly Learning Synthetic Data Generation ; Simulation is increasingly being used for generating large labelled datasets in many machine learning problems. Recent methods have focused on adjusting simulator parameters with the goal of maximising accuracy on a validation task, usually relying on REINFORCElike gradient estimators. However these approaches are very expensive as they treat the entire data generation, model training, and validation pipeline as a blackbox and require multiple costly objective evaluations at each iteration. We propose an efficient alternative for optimal synthetic data generation, based on a novel differentiable approximation of the objective. This allows us to optimize the simulator, which may be nondifferentiable, requiring only one objective evaluation at each iteration with a little overhead. We demonstrate on a stateoftheart photorealistic renderer that the proposed method finds the optimal data distribution faster up to 50times, with significantly reduced training data generation up to 30times and better accuracy 8.7 on realworld test datasets than previous methods.

Quantum phenomenological gravitational dynamics A general view from thermodynamics of spacetime ; In this work we derive general quantum phenomenological equations of gravitational dynamics and analyse its features. The derivation uses the formalism developed in thermodynamics of spacetime and introduces low energy quantum gravity modifications to it. Quantum gravity effects are considered via modification of Bekenstein entropy by an extra logarithmic term in the area. This modification is predicted by several approaches to quantum gravity, including loop quantum gravity, string theory, AdSCFT correspondence and generalised uncertainty principle phenomenology, giving our result a general character. The derived equations generalise classical equations of motion of unimodular gravity, instead of the ones of general relativity, and they contain at most second derivatives of the metric. We provide two independent derivations of the equations based on thermodynamics of local causal diamonds. First one uses Jacobson's maximal vacuum entanglement hypothesis, the second one Clausius entropy flux. Furthermore, we consider questions of diffeomorphism and local Lorentz invariance of the resulting dynamics and discuss its application to a simple cosmological model, finding a resolution of the classical singularity.

Impact of Leakage for Electricity Generation by Pyroelectric Converter ; Pyroelectric energy converter is a functional capacitor using pyroelectric material as the dielectric layer. Utilizing the firstorder phase transformation of the material, the pyroelectric device can generate adequate electricity within small temperature fluctuations. However, most pyroelectric capacitors are leaking during energy conversion. In this paper, we analyze the thermodynamics of pyroelectric energy conversion with consideration of the electric leakage. Our thermodynamic model is verified by experiments using three phasetransforming ferroelectric materials with different pyroelectric properties and leakage behaviors. We demonstrate that the impact of leakage for electric generation is prominent, and sometimes may be confused with the actual power generation by pyroelectricity. We discover an ideal material candidate, Ba,CaTi,Zr,CeO3, which exhibits large pyroelectric current and extremely low leakage current. The pyroelectric converter made of this material generates 1.95 muAcm2 pyroelectric current density and 0.2 Jcm3 pyroelectric work density even after 1389 thermodynamic conversion cycles.

Boosting Retailer Revenue by Generated Optimized Combined Multiple Digital Marketing Campaigns ; Campaign is a frequently employed instrument in lifting up the GMV Gross Merchandise Volume of retailer in traditional marketing. As its counterpart in online context, digitalmarketingcampaign DMC has being trending in recent years with the rapid development of the ecommerce. However, how to empower massive sellers on the online retailing platform the capacity of applying combined multiple digital marketing campaigns to boost their shops' revenue, is still a novel topic. In this work, a comprehensive solution of generating optimized combined multiple DMCs is presented. Firstly, a potential personalized DMC pool is generated for every retailer by a newly proposed neural network model, i.e. the DMCNet DigitalMarketingCampaign Net. Secondly, based on the submodular optimization theory and the DMC pool by DMCNet, the generated combined multiple DMCs are ranked with respect to their revenue generation strength then the top three ranked campaigns are returned to the sellers' backend management system, so that retailers can set combined multiple DMCs for their online shops just in oneshot. Real online ABtest shows that with the integrated solution, sellers of the online retailing platform increase their shops' GMVs with approximately 6.

Interdatabase validation of a deep learning approach for automatic sleep scoring ; In this work we describe a new deep learning approach for automatic sleep staging, and carry out its validation by addressing its generalization capabilities on a wide range of sleep staging databases. Prediction capabilities are evaluated in the context of independent local and external generalization scenarios. Effectively, by comparing both procedures it is possible to better extrapolate the expected performance of the method on the general reference task of sleep staging, regardless of data from a specific database. In addition, we examine the suitability of a novel approach based on the use of an ensemble of individual local models and evaluate its impact on the resulting interdatabase generalization performance. Validation results show good general performance, as compared to the expected levels of human expert agreement, as well as stateoftheart automatic sleep staging approaches

SGD Generalizes Better Than GD And Regularization Doesn't Help ; We give a new separation result between the generalization performance of stochastic gradient descent SGD and of fullbatch gradient descent GD in the fundamental stochastic convex optimization model. While for SGD it is wellknown that O1epsilon2 iterations suffice for obtaining a solution with epsilon excess expected risk, we show that with the same number of steps GD may overfit and emit a solution with Omega1 generalization error. Moreover, we show that in fact Omega1epsilon4 iterations are necessary for GD to match the generalization performance of SGD, which is also tight due to recent work by Bassily et al. 2020. We further discuss how regularizing the empirical risk minimized by GD essentially does not change the above result, and revisit the concepts of stability, implicit bias and the role of the learning algorithm in generalization.

Recurrent Neural Network for MoonBoard Climbing Route Classification and Generation ; Classifying the difficulties of climbing routes and generating new routes are both challenging. Existing machine learning models not only fail to accurately predict a problem's difficulty, but they are also unable to generate reasonable problems. In this work, we introduced BetaMove, a new move preprocessing pipeline we developed, in order to mimic a human climber's hand sequence. The preprocessed move sequences were then used to train both a route generator and a grade predictor. By preprocessing a MoonBoard problem into a proper move sequence, the accuracy of our grade predictor reaches near humanlevel performance, and our route generator produces new routes of much better quality compared to previous work. We demonstrated that with BetaMove, we are able to inject human insights into the machine learning problems, and this can be the foundations for future transfer learning on climbing style classification problems.

Transductive ZeroShot Learning by Decoupled Feature Generation ; In this paper, we address zeroshot learning ZSL, the problem of recognizing categories for which no labeled visual data are available during training. We focus on the transductive setting, in which unlabelled visual data from unseen classes is available. Stateoftheart paradigms in ZSL typically exploit generative adversarial networks to synthesize visual features from semantic attributes. We posit that the main limitation of these approaches is to adopt a single model to face two problems 1 generating realistic visual features, and 2 translating semantic attributes into visual cues. Differently, we propose to decouple such tasks, solving them separately. In particular, we train an unconditional generator to solely capture the complexity of the distribution of visual data and we subsequently pair it with a conditional generator devoted to enrich the prior knowledge of the data distribution with the semantic content of the class embeddings. We present a detailed ablation study to dissect the effect of our proposed decoupling approach, while demonstrating its superiority over the related stateoftheart.

Cultureinspired Multimodal Color Palette Generation and Colorization A Chinese Youth Subculture Case ; Color is an essential component of graphic design, acting not only as a visual factor but also carrying cultural implications. However, existing research on algorithmic color palette generation and colorization largely ignores the cultural aspect. In this paper, we contribute to this line of research by first constructing a unique color dataset inspired by a specific culture, i.e., Chinese Youth Subculture CYS, which is an vibrant and trending cultural group especially for the Gen Z population. We show that the colors used in CYS have special aesthetic and semantic characteristics that are different from generic color theory. We then develop an interactive multimodal generative framework to create CYSstyled color palettes, which can be used to put a CYS twist on images using our automatic colorization model. Our framework is illustrated via a demo system designed with the humanintheloop principle that constantly provides feedback to our algorithms. User studies are also conducted to evaluate our generation results.

Conditional Loss and Deep Euler Scheme for Time Series Generation ; We introduce three new generative models for time series that are based on Euler discretization of Stochastic Differential Equations SDEs and Wasserstein metrics. Two of these methods rely on the adaptation of generative adversarial networks GANs to time series. The third algorithm, called Conditional Euler Generator CEGEN, minimizes a dedicated distance between the transition probability distributions over all time steps. In the context of Ito processes, we provide theoretical guarantees that minimizing this criterion implies accurate estimations of the drift and volatility parameters. We demonstrate empirically that CEGEN outperforms stateoftheart and GAN generators on both marginal and temporal dynamics metrics. Besides, it identifies accurate correlation structures in high dimension. When few data points are available, we verify the effectiveness of CEGEN, when combined with transfer learning methods on Monte Carlo simulations. Finally, we illustrate the robustness of our method on various realworld datasets.

Largescale GHZ states through topologically protected zeroenergy mode in a superconducting qutritresonator chain ; We propose a superconducting qutritresonator chain model, and analytically work out forms of its topological edge states. The existence of the zeroenergy mode enables to generate a state transfer between two ends of the chain, accompanied with state flips of all intermediate qutrits, based on which Nbody GreenbergerHorneZeilinger GHZ states can be generated with great robustness against disorders of coupling strengths. Three schemes of generating largescale GHZ states are designed, each of which possesses the robustness against loss of qutrits or of resonators, meeting a certain performance requirement of different experimental devices. With experimentally feasible qutritresonator coupling strengths and available coherence times of qutrits and resonators, it has a potential to generate largescale GHZ states among dozens of qutrits with a high fidelity. Further, we show the experimental consideration of generating GHZ states based on the circuit QED system, and discuss the prospect of realizing fast GHZ states.

Generalizing Decision Making for Automated Driving with an Invariant Environment Representation using Deep Reinforcement Learning ; Data driven approaches for decision making applied to automated driving require appropriate generalization strategies, to ensure applicability to the world's variability. Current approaches either do not generalize well beyond the training data or are not capable to consider a variable number of traffic participants. Therefore we propose an invariant environment representation from the perspective of the ego vehicle. The representation encodes all necessary information for safe decision making. To assess the generalization capabilities of the novel environment representation, we train our agents on a small subset of scenarios and evaluate on the entire diverse set of scenarios. Here we show that the agents are capable to generalize successfully to unseen scenarios, due to the abstraction. In addition we present a simple occlusion model that enables our agents to navigate intersections with occlusions without a significant change in performance.

ASAM Adaptive SharpnessAware Minimization for ScaleInvariant Learning of Deep Neural Networks ; Recently, learning algorithms motivated from sharpness of loss surface as an effective measure of generalization gap have shown stateoftheart performances. Nevertheless, sharpness defined in a rigid region with a fixed radius, has a drawback in sensitivity to parameter rescaling which leaves the loss unaffected, leading to weakening of the connection between sharpness and generalization gap. In this paper, we introduce the concept of adaptive sharpness which is scaleinvariant and propose the corresponding generalization bound. We suggest a novel learning method, adaptive sharpnessaware minimization ASAM, utilizing the proposed generalization bound. Experimental results in various benchmark datasets show that ASAM contributes to significant improvement of model generalization performance.

False Relay Operation Attacks in Power Systems with High Renewables ; Loadgeneration balance and system inertia are essential for maintaining frequency in power systems. Power grids are equipped with RateofChangeof Frequency ROCOF and Load Shedding LS relays in order to keep loadgeneration balance. With the increasing penetration of renewables, the inertia of the power grids is declining, which results in a faster drop in system frequency in case of loadgeneration imbalance. In this context, we analyze the feasibility of launching False Data Injection FDI in order to create False Relay Operations FRO, which we refer to as FRO attack, in the power systems with high renewables. We model the frequency dynamics of the power systems and corresponding FDI attacks, including the impact of parameters, such as synchronous generators inertia, and governors time constant and droop, on the success of FRO attacks. We formalize the FRO attack as a Constraint Satisfaction Problem CSP and solve using Satisfiability Modulo Theories SMT. Our case studies show that power grids with renewables are more susceptible to FRO attacks and the inertia of synchronous generators plays a critical role in reducing the success of FRO attacks in the power grids.

Can Transformers Jump Around Right in Natural Language Assessing Performance Transfer from SCAN ; Despite their practical success, modern seq2seq architectures are unable to generalize systematically on several SCAN tasks. Hence, it is not clear if SCANstyle compositional generalization is useful in realistic NLP tasks. In this work, we study the benefit that such compositionality brings about to several machine translation tasks. We present several focused modifications of Transformer that greatly improve generalization capabilities on SCAN and select one that remains on par with a vanilla Transformer on a standard machine translation MT task. Next, we study its performance in lowresource settings and on a newly introduced distributionshifted EnglishFrench translation task. Overall, we find that improvements of a SCANcapable model do not directly transfer to the resourcerich MT setup. In contrast, in the lowresource setup, general modifications lead to an improvement of up to 13.1 BLEU score w.r.t. a vanilla Transformer. Similarly, an improvement of 14 in an accuracybased metric is achieved in the introduced compositional EnglishFrench translation task. This provides experimental evidence that the compositional generalization assessed in SCAN is particularly useful in resourcestarved and domainshifted scenarios.

Grid Partitioned Attention Efficient TransformerApproximation with Inductive Bias for High Resolution Detail Generation ; Attention is a general reasoning mechanism than can flexibly deal with image information, but its memory requirements had made it so far impractical for high resolution image generation. We present Grid Partitioned Attention GPA, a new approximate attention algorithm that leverages a sparse inductive bias for higher computational and memory efficiency in image domains queries attend only to few keys, spatially close queries attend to close keys due to correlations. Our paper introduces the new attention layer, analyzes its complexity and how the tradeoff between memory usage and model power can be tuned by the hyperparameters.We will show how such attention enables novel deep learning architectures with copying modules that are especially useful for conditional image generation tasks like pose morphing. Our contributions are i algorithm and code1of the novel GPA layer, ii a novel deep attentioncopying architecture, and iii new stateofthe art experimental results in human pose morphing generation benchmarks.

WhiteBox Cartoonization Using An Extended GAN Framework ; In the present study, we propose to implement a new framework for estimating generative models via an adversarial process to extend an existing GAN framework and develop a whitebox controllable image cartoonization, which can generate highquality cartooned imagesvideos from realworld photos and videos. The learning purposes of our system are based on three distinct representations surface representation, structure representation, and texture representation. The surface representation refers to the smooth surface of the images. The structure representation relates to the sparse colour blocks and compresses generic content. The texture representation shows the texture, curves, and features in cartoon images. Generative Adversarial Network GAN framework decomposes the images into different representations and learns from them to generate cartoon images. This decomposition makes the framework more controllable and flexible which allows users to make changes based on the required output. This approach overcomes any previous system in terms of maintaining clarity, colours, textures, shapes of images yet showing the characteristics of cartoon images.

SelfSupervised Generative Adversarial Network for Depth Estimation in Laparoscopic Images ; Dense depth estimation and 3D reconstruction of a surgical scene are crucial steps in computer assisted surgery. Recent work has shown that depth estimation from a stereo images pair could be solved with convolutional neural networks. However, most recent depth estimation models were trained on datasets with perpixel ground truth. Such data is especially rare for laparoscopic imaging, making it hard to apply supervised depth estimation to real surgical applications. To overcome this limitation, we propose SADepth, a new selfsupervised depth estimation method based on Generative Adversarial Networks. It consists of an encoderdecoder generator and a discriminator to incorporate geometry constraints during training. Multiscale outputs from the generator help to solve the local minima caused by the photometric reprojection loss, while the adversarial learning improves the framework generation quality. Extensive experiments on two public datasets show that SADepth outperforms recent stateoftheart unsupervised methods by a large margin, and reduces the gap between supervised and unsupervised depth estimation in laparoscopic images.

Between Flexibility and Consistency Joint Generation of Captions and Subtitles ; Speech translation ST has lately received growing interest for the generation of subtitles without the need for an intermediate source language transcription and timing i.e. captions. However, the joint generation of source captions and target subtitles does not only bring potential output quality advantages when the two decoding processes inform each other, but it is also often required in multilingual scenarios. In this work, we focus on ST models which generate consistent captionssubtitles in terms of structure and lexical content. We further introduce new metrics for evaluating subtitling consistency. Our findings show that joint decoding leads to increased performance and consistency between the generated captions and subtitles while still allowing for sufficient flexibility to produce subtitles conforming to languagespecific needs and norms.

Generation of Photon Pairs by Stimulated Emission in Ring Resonators ; Thirdorder parametric downconversion TOPDC describes a class of nonlinear interactions in which a pump photon is converted into a photon triplet. This process can occur spontaneously, or it can be stimulated by seeding fields. In the former case, one typically has the generation of nonGaussian states of light. In the latter, the situation is more variegated, for stimulated TOPDC StTOPDC can be implemented in many ways, depending on the number and properties of the seeding fields. Here we show that StTOPDC can be exploited for the generation of quantum correlated photon pairs. We examine the peculiar features of this approach when compared with secondorder spontaneous parametric downconversion and spontaneous fourwave mixing. We model StTOPDC in a microring resonator, predicting observable generation rates in a microring engineered for thirdharmonic generation. We conclude that if the experimental difficulties associated with implementing StTOPDC can be overcome, it may soon be possible to demonstrate this process in resonant integrated devices.