Datasets:

sswt

/

arxiv_sample_50K

like 0

FiniteSize Scaling of a FirstOrder Dynamical Phase Transition Adaptive Population Dynamics and an Effective Model ; We analyze large deviations of the timeaveraged activity in the one dimensional FredricksonAndersen model, both numerically and analytically. The model exhibits a dynamical phase transition, which appears as a singularity in the large deviation function. We analyze the finitesize scaling of this phase transition numerically, by generalizing an existing cloning algorithm to include a multicanonical feedback control this significantly improves the computational efficiency. Motivated by these numerical results, we formulate an effective theory for the model in the vicinity of the phase transition, which accounts quantitatively for the observed behavior. We discuss potential applications of the numerical method and the effective theory in a range of more general contexts.

Detecting gravitational decoherence with clocks Limits on temporal resolution from a classical channel model of gravity ; The notion of time is given a different footing in Quantum Mechanics and General Relativity, treated as a parameter in the former and being an observer dependent property in the later. From a operational point of view time is simply the correlation between a system and a clock, where an idealized clock can be modelled as a two level systems. We investigate the dynamics of clocks interacting gravitationally by treating the gravitational interaction as a classical information channel. In particular, we focus on the decoherence rates and temporal resolution of arrays of N clocks showing how the minimum dephasing rate scales with N, and the spatial configuration. Furthermore, we consider the gravitational redshift between a clock and massive particle and show that a classical channel model of gravity predicts a finite dephasing rate from the nonlocal interaction. In our model we obtain a fundamental limitation in time accuracy that is intrinsic to each clock.

Endtoend Learning for 3D Facial Animation from Raw Waveforms of Speech ; We present a deep learning framework for realtime speechdriven 3D facial animation from just raw waveforms. Our deep neural network directly maps an input sequence of speech audio to a series of micro facial action unit activations and head rotations to drive a 3D blendshape face model. In particular, our deep model is able to learn the latent representations of timevarying contextual information and affective states within the speech. Hence, our model not only activates appropriate facial action units at inference to depict different utterance generating actions, in the form of lip movements, but also, without any assumption, automatically estimates emotional intensity of the speaker and reproduces her everchanging affective states by adjusting strength of facial unit activations. For example, in a happy speech, the mouth opens wider than normal, while other facial units are relaxed; or in a surprised state, both eyebrows raise higher. Experiments on a diverse audiovisual corpus of different actors across a wide range of emotional states show interesting and promising results of our approach. Being speakerindependent, our generalized model is readily applicable to various tasks in humanmachine interaction and animation.

A Semantic Relevance Based Neural Network for Text Summarization and Text Simplification ; Text summarization and text simplification are two major ways to simplify the text for poor readers, including children, nonnative speakers, and the functionally illiterate. Text summarization is to produce a brief summary of the main ideas of the text, while text simplification aims to reduce the linguistic complexity of the text and retain the original meaning. Recently, most approaches for text summarization and text simplification are based on the sequencetosequence model, which achieves much success in many text generation tasks. However, although the generated simplified texts are similar to source texts literally, they have low semantic relevance. In this work, our goal is to improve semantic relevance between source texts and simplified texts for text summarization and text simplification. We introduce a Semantic Relevance Based neural model to encourage high semantic similarity between texts and summaries. In our model, the source text is represented by a gated attention encoder, while the summary representation is produced by a decoder. Besides, the similarity score between the representations is maximized during training. Our experiments show that the proposed model outperforms the stateoftheart systems on two benchmark corpus.

Describing Natural Images Containing Novel Objects with Knowledge Guided Assitance ; Images in the wild encapsulate rich knowledge about varied abstract concepts and cannot be sufficiently described with models built only using imagecaption pairs containing selected objects. We propose to handle such a task with the guidance of a knowledge base that incorporate many abstract concepts. Our method is a twostep process where we first build a multientitylabel image recognition model to predict abstract concepts as image labels and then leverage them in the second step as an external semantic attention and constrained inference in the caption generation model for describing images that depict unseennovel objects. Evaluations show that our models outperform most of the prior work for outofdomain captioning on MSCOCO and are useful for integration of knowledge and vision in general.

Subtractive Color Mixture Computation ; Modeling subtractive color mixture e.g., the way that paints mix is difficult when working with colors described only by threedimensional color space values, such as RGB. Although RGB values are sufficient to describe a specific color sensation, they do not contain enough information to predict the RGB color that would result from a subtractive mixture of two specified RGB colors. Methods do exist for accurately modeling subtractive mixture, such as the KubelkaMunk equations, but require extensive spectrophotometric measurements of the mixed components, making them unsuitable for many computer graphics applications. This paper presents a strategy for modeling subtractive color mixture given only the RGB information of the colors being mixed, written for a general audience. The RGB colors are first transformed to generic, representative spectral distributions, and then this spectral information is used to perform the subtractive mixture, using the weighted arithmeticgeometric mean. This strategy provides reasonable, representative subtractive mixture colors with only modest computational effort and no experimental measurements. As such, it provides a useful way to model subtractive color mixture in computer graphics applications.

Generalized linear mixing model accounting for endmember variability ; Endmember variability is an important factor for accurately unveiling vital information relating the pure materials and their distribution in hyperspectral images. Recently, the extended linear mixing model ELMM has been proposed as a modification of the linear mixing model LMM to consider endmember variability effects resulting mainly from illumination changes. In this paper, we further generalize the ELMM leading to a new model GLMM to account for more complex spectral distortions where different wavelength intervals can be affected unevenly. We also extend the existing methodology to jointly estimate the variability and the abundances for the GLMM. Simulations with real and synthetic data show that the unmixing process can benefit from the extra flexibility introduced by the GLMM.

Quasilocal charges and the Generalized Gibbs Ensemble in the LiebLiniger model ; We consider the construction of a generalized Gibbs ensemble composed of complete bases of conserved charges in the repulsive LiebLiniger model. We will show that it is possible to construct these bases with varying locality as well as demonstrating that such constructions are always possible provided one has in hand at least one complete basis set of charges. This procedure enables the construction of bases of charges that possess well defined, finite expectation values given an arbitrary initial state. We demonstrate the use of these charges in the context of two different quantum quenches a quench where the strength of the interactions in a onedimensional gas is switched suddenly from zero to some finite value and the release of a one dimensional cold atomic gas from a confining parabolic trap. While we focus on the LiebLiniger model in this paper, the principle of the construction of these charges applies to all integrable models, both in continuum and lattice form.

Nebula F0 Estimation and Voicing Detection by Modeling the Statistical Properties of Feature Extractors ; A F0 and voicing status estimation algorithm for high quality speech analysissynthesis is proposed. This problem is approached from a different perspective that models the behavior of feature extractors under noise, instead of directly modeling speech signals. Under timefrequency locality assumptions, the joint distribution of extracted features and target F0 can be characterized by training a bank of Gaussian mixture models GMM on artificial data generated from MonteCarlo simulations. The trained GMMs can then be used to generate a set of conditional distributions on the predicted F0, which are then combined and postprocessed by Viterbi algorithm to give a final F0 trajectory. Evaluation on CSTR and CMU Arctic speech databases shows that the proposed method, trained on fully synthetic data, achieves lower gross error rates than stateoftheart methods.

Classical and Quantum Aspects of YangBaxter WessZumino Models ; We investigate the integrable YangBaxter deformation of the 2d Principal Chiral Model with a WessZumino term. For arbitrary groups, the oneloop beta functions are calculated and display a surprising connection between classical and quantum physics the classical integrability condition is necessary to prevent new couplings being generated by renormalisation. We show these theories admit an elegant realisation of PoissonLie Tduality acting as a simple inversion of coupling constants. The selfdual point corresponds to the WessZuminoWitten model and is the IR fixed point under RG. We address the possibility of having supersymmetric extensions of these models showing that extended supersymmetry is not possible in general.

Building Datadriven Models with Microstructural Images Generalization and Interpretability ; As datadriven methods rise in popularity in materials science applications, a key question is how these machine learning models can be used to understand microstructure. Given the importance of processstructureproperty relations throughout materials science, it seems logical that models that can leverage microstructural data would be more capable of predicting property information. While there have been some recent attempts to use convolutional neural networks to understand microstructural images, these early studies have focused only on which featurizations yield the highest machine learning model accuracy for a single data set. This paper explores the use of convolutional neural networks for classifying microstructure with a more holistic set of objectives in mind generalization between data sets, number of features required, and interpretability.

Marginal Loglinear Parameters and their Collapsibility for Categorical Data ; We consider marginal loglinear models for parameterizing distributions on multidimensional contingency tables. These models generalize ordinary loglinear and multivariate logistic models, besides several others. First, we obtain some characteristic properties of marginal loglinear parameters. Then we define collapsibility and strict collapsibility of these parameters in a general sense. Several necessary and sufficient conditions for collapsibility and strict collapsibility are derived based on simple functions of only the cell probabilities, which are easily verifiable. These include results for an arbitrary set of marginal loglinear parameters having some common effects. The connections of strict collapsibility to various forms of independence of the variables are explored. We analyze some reallife datasets to illustrate the above results on collapsibility and strict collapsibility. Finally, we obtain a result relating parameters with the same effect but different margins for an arbitrary table, and demonstrate smoothness of marginal loglinear models under collapsibility conditions.

Neural Discrete Representation Learning ; Learning useful representations without supervision remains a key challenge in machine learning. In this paper, we propose a simple yet powerful generative model that learns such discrete representations. Our model, the Vector QuantisedVariational AutoEncoder VQVAE, differs from VAEs in two key ways the encoder network outputs discrete, rather than continuous, codes; and the prior is learnt rather than static. In order to learn a discrete latent representation, we incorporate ideas from vector quantisation VQ. Using the VQ method allows the model to circumvent issues of posterior collapse where the latents are ignored when they are paired with a powerful autoregressive decoder typically observed in the VAE framework. Pairing these representations with an autoregressive prior, the model can generate high quality images, videos, and speech as well as doing high quality speaker conversion and unsupervised learning of phonemes, providing further evidence of the utility of the learnt representations.

Foreaft asymmetric flocking ; We show that foreaft asymmetry, a generic feature of living organisms and some active matter systems, can have a strong influence on the collective properties of even the simplest flocking models. Specifically, an arbitrarily weak asymmetry favoring front neighbors changes qualitatively the phase diagram of the Vicsek model. A region where many sharp traveling band solutions coexist is present at low noise strength, below the TonerTu liquid, at odds with the phaseseparation scenario well describing the usual isotropic model. Inside this region, a banded liquid' phase with algebraic density distribution coexists with band solutions. Linear stability analysis at the hydrodynamic level suggests that these results are generic and not specific to the Vicsek model.

Codes for Correcting Localized Deletions ; We consider the problem of constructing binary codes for correcting deletions that are localized within certain parts of the codeword that are unknown a priori. The model that we study is when delta leq w deletions are localized in a window of size w bits. These delta deletions do not necessarily occur in consecutive positions, but are restricted to the window of size w. The localized deletions model is a generalization of the bursty model, in which all the deleted bits are consecutive. In this paper, we construct new explicit codes for the localized model, based on the family of Guess Check codes which was previously introduced by the authors. The codes that we construct can correct, with high probability, delta leq w deletions that are localized in a single window of size w, where w grows with the block length. Moreover, these codes are systematic; have low redundancy; and have efficient deterministic encoding and decoding algorithms. We also generalize these codes to deletions that are localized within multiple windows in the codeword.

Evidence Aggregation for Answer ReRanking in OpenDomain Question Answering ; A popular recent approach to answering opendomain questions is to first search for questionrelated passages and then apply reading comprehension models to extract answers. Existing methods usually extract answers from single passages independently. But some questions require a combination of evidence from across different sources to answer correctly. In this paper, we propose two models which make use of multiple passages to generate their answers. Both use an answerreranking approach which reorders the answer candidates generated by an existing stateoftheart QA model. We propose two methods, namely, strengthbased reranking and coveragebased reranking, to make use of the aggregated evidence from different passages to better determine the answer. Our models have achieved stateoftheart results on three public opendomain QA datasets QuasarT, SearchQA and the opendomain version of TriviaQA, with about 8 percentage points of improvement over the former two datasets.

Fast and reliable inference algorithm for hierarchical stochastic block models ; Network clustering reveals the organization of a network or corresponding complex system with elements represented as vertices and interactions as edges in a directed, weighted graph. Although the notion of clustering can be somewhat loose, network clusters or groups are generally considered as nodes with enriched interactions and edges sharing common patterns. Statistical inference often treats groups as latent variables, with observed networks generated from latent group structure, termed a stochastic block model. Regardless of the definitions, statistical inference can be either translated to modularity maximization, which is provably an NPcomplete problem. Here we present scalable and reliable algorithms that recover hierarchical stochastic block models fast and accurately. Our algorithm scales almost linearly in number of edges, and inferred models were more accurate that other scalable methods.

A Generally Applicable, Highly Scalable Measurement Computation and Optimization Approach to Sequential ModelBased Diagnosis ; ModelBased Diagnosis deals with the identification of the real cause of a system's malfunction based on a formal system model and observations of the system behavior. When a malfunction is detected, there is usually not enough information available to pinpoint the real cause and one needs to discriminate between multiple fault hypotheses called diagnoses. To this end, Sequential Diagnosis approaches ask an oracle for additional system measurements. This work presents strategies for optimal measurement selection in modelbased sequential diagnosis. In particular, assuming a set of leading diagnoses being given, we show how queries sets of measurements can be computed and optimized along two dimensions expected number of queries and cost per query. By means of a suitable decoupling of two optimizations and a clever search space reduction the computations are done without any inference engine calls. For the full search space, we give a method requiring only a polynomial number of inferences and show how query properties can be guaranteed which existing methods do not provide. Evaluation results using realworld problems indicate that the new method computes virtually optimal queries instantly independently of the size and complexity of the considered diagnosis problems and outperforms equally general methods not exploiting the proposed theory by orders of magnitude.

Attentive Explanations Justifying Decisions and Pointing to the Evidence Extended Abstract ; Deep models are the defacto standard in visual decision problems due to their impressive performance on a wide array of visual tasks. On the other hand, their opaqueness has led to a surge of interest in explainable systems. In this work, we emphasize the importance of model explanation in various forms such as visual pointing and textual justification. The lack of data with justification annotations is one of the bottlenecks of generating multimodal explanations. Thus, we propose two largescale datasets with annotations that visually and textually justify a classification decision for various activities, i.e. ACTX, and for question answering, i.e. VQAX. We also introduce a multimodal methodology for generating visual and textual explanations simultaneously. We quantitatively show that training with the textual explanations not only yields better textual justification models, but also models that better localize the evidence that support their decision.

A Nakanishibased model illustrating the covariant extension of the pion GPD overlap representation and its ambiguities ; A systematic approach for the model building of Generalized Parton Distributions GPDs, based on their overlap representation within the DGLAP kinematic region and a further covariant extension to the ERBL one, is applied to the valencequark pion's case, using lightfront wave functions inspired by the Nakanishi representation of the pion's BetheSalpeter amplitudes BSA. This simple but fruitful pion's GPD model illustrates the general model building technique and, in addition, allows for the ambiguities related to the covariant extension, grounded on the Double Distribution DD representation, to be constrained by requiring a softpion theorem to be properly observed.

PredictionConstrained Topic Models for Antidepressant Recommendation ; Supervisory signals can help topic models discover lowdimensional data representations that are more interpretable for clinical tasks. We propose a framework for training supervised latent Dirichlet allocation that balances two goals faithful generative explanations of highdimensional data and accurate prediction of associated class labels. Existing approaches fail to balance these goals by not properly handling a fundamental asymmetry the intended task is always predicting labels from data, not data from labels. Our new predictionconstrained objective trains models that predict labels from heldout data well while also producing good generative likelihoods and interpretable topicword parameters. In a case study on predicting depression medications from electronic health records, we demonstrate improved recommendations compared to previous supervised topic models and high dimensional logistic regression from words alone.

Recurrent Neural Networks for Semantic Instance Segmentation ; We present a recurrent model for semantic instance segmentation that sequentially generates binary masks and their associated class probabilities for every object in an image. Our proposed system is trainable endtoend from an input image to a sequence of labeled masks and, compared to methods relying on object proposals, does not require postprocessing steps on its output. We study the suitability of our recurrent model on three different instance segmentation benchmarks, namely Pascal VOC 2012, CVPPP Plant Leaf Segmentation and Cityscapes. Further, we analyze the object sorting patterns generated by our model and observe that it learns to follow a consistent pattern, which correlates with the activations learned in the encoder part of our network. Source code and models are available at httpsimatgeupc.github.iorsis

Generative Adversarial Perturbations ; In this paper, we propose novel generative models for creating adversarial examples, slightly perturbed images resembling natural images but maliciously crafted to fool pretrained models. We present trainable deep neural networks for transforming images to adversarial perturbations. Our proposed models can produce imageagnostic and imagedependent perturbations for both targeted and nontargeted attacks. We also demonstrate that similar architectures can achieve impressive results in fooling classification and semantic segmentation models, obviating the need for handcrafting attack methods for each task. Using extensive experiments on challenging highresolution datasets such as ImageNet and Cityscapes, we show that our perturbations achieve high fooling rates with small perturbation norms. Moreover, our attacks are considerably faster than current iterative methods at inference time.

Axions and ALPs a very short introduction ; The QCD axion was originally predicted as a dynamical solution to the strong CP problem. Axion like particles ALPs are also a generic prediction of many high energy physics models including string theory. Theoretical models for axions are reviewed, giving a generic multiaxion action with couplings to the standard model. The couplings and masses of these axions can span many orders of magnitude, and cosmology leads us to consider several distinct populations of axions behaving as coherent condensates, or relativistic particles. Light, stable axions are a mainstay dark matter candidate. Axion cosmology and calculation of the relic density are reviewed. A very brief survey is given of the phenomenology of axions arising from their direct couplings to the standard model, and their distinctive gravitational interactions.

Spin polarization through Floquet resonances in a driven central spin model ; Adiabatically varying the driving frequency of a periodicallydriven manybody quantum system can induce controlled transitions between resonant eigenstates of the timeaveraged Hamiltonian, corresponding to adiabatic transitions in the Floquet spectrum and presenting a general tool in quantum manybody control. Using the central spin model as an application, we show how such controlled driving processes can lead to a polarizationbased decoupling of the central spin from its decoherenceinducing environment at resonance. While it is generally impossible to obtain the exact Floquet Hamiltonian in driven interacting systems, we exploit the integrability of the central spin model to show how techniques from quantum quenches can be used to explicitly construct the Floquet Hamiltonian in a restricted manybody basis and model Floquet resonances.

Pretraining Attention Mechanisms ; Recurrent neural networks with differentiable attention mechanisms have had success in generative and classification tasks. We show that the classification performance of such models can be enhanced by guiding a randomly initialized model to attend to salient regions of the input in early training iterations. We further show that, if explicit heuristics for guidance are unavailable, a model that is pretrained on an unsupervised reconstruction task can discover good attention policies without supervision. We demonstrate that increased efficiency of the attention mechanism itself contributes to these performance improvements. Based on these insights, we introduce bootstrapped glimpse mimicking, a simple, theoretically taskgeneral method of more effectively training attention models. Our work draws inspiration from and parallels results on human learning of attention.

Inferring hierarchical structure of spatial and generic complex networks through a modeling framework ; Our recent paper Grauwin et al. Sci. Rep. 7 2017 demonstrates that community and hierarchical structure of the networks of human interactions largely determines the least and should be taken into account while modeling them. In the present proofofconcept preprint the opposite question is considered could the hierarchical structure itself be inferred to be best aligned with the network model The inference mechanism is provided for both spatial networks as well as complex networks in general through a model based on hierarchical and if defined geographical distances. The mechanism allows to discover hierarchical and community structure at any desired resolution in complex networks and in particular the spaceindependent structure of the spatial networks. The approach is illustrated on the example of the interstate people migration network in USA.

Improvements to Inference Compilation for Probabilistic Programming in LargeScale Scientific Simulators ; We consider the problem of Bayesian inference in the family of probabilistic models implicitly defined by stochastic generative models of data. In scientific fields ranging from population biology to cosmology, lowlevel mechanistic components are composed to create complex generative models. These models lead to intractable likelihoods and are typically nondifferentiable, which poses challenges for traditional approaches to inference. We extend previous work in inference compilation, which combines universal probabilistic programming and deep learning methods, to largescale scientific simulators, and introduce a C based probabilistic programming library called CPProb. We successfully use CPProb to interface with SHERPA, a large codebase used in particle physics. Here we describe the technical innovations realized and planned for this library.

Cosmological Analysis of Reconstructed mathcalFT,TmathcalG Models ; In this paper, we analyze cosmological consequences of the reconstructed generalized ghost pilgrim dark energy mathcalFT,TmathcalG models in terms of redshift parameter z. For this purpose, we consider powerlaw scale factor, scale factor for two unified phases and intermediate scale factor. We discuss graphical behavior of the reconstructed models and examine their stability analysis. Also, we explore the behavior of equation of state as well as deceleration parameters and omegaLambdaomegaLambda' as well as rs planes. It is found that all models are stable for pilgrim dark energy parameter 2. The equation of state parameter satisfies the necessary condition for pilgrim dark energy phenomenon for all scale factors. All other cosmological parameters show great consistency with the current behavior of the universe.

Asymptotic behavior for a nonautonomous model of neural fields with variable external stimulus ; In this work we consider a class of nonlocal nonautonomous evolution equations, which generalizes the model of neuronal activity that arises in Amari 1979. Under suitable assumptions on the nonlinearity and on the parameters present in the equation, we study, in an appropriated Banach space, the assimptotic behavior of the evolution process generated by this equation. We prove results on existence, uniqueness and smoothness of the solutions and on the existence of pullback attracts for the evolution process associated to this equation. We also prove a continuous dependence of the evolution process with respect to external stimulus function present in the model. Furthermore, using the result of continuous dependence of the evolution process, we also prove the upper semicontinuity of pullback attracts with respect to stimulus function. We conclude with a small discussion about the model and about a biological interpretation of the result of continuous dependence of neuronal activity with respect to the external stimulus function.

Charged VaidyaTikekar model for super compact star ; In this work, we explore a class of compact charged spheres that have been tested against experimental and observational constraints with some known compact stars candidates. The study is performed by considering the selfgravitating, charged, isotropic fluids which is more pliability in solving the EinsteinMaxwell equations. In order to determine the interior geometry, we utilize the VaidyaTikekar geometry for the metric potential with RiessnerNordstrom metric as an exterior solution. In this models, we determine constants after selecting some particular values of M and R, for the compact objects SAX J1808.43658, Her X1 and 4U 153852. The most striking consequence is that hydrostatic equilibrium is maintained for different forces, and the situation is clarified by using the generalized TolmanOppenheimerVolkoff TOV equation. In addition to this, we also present the energy conditions, speeds of sound and compactness of stars that are very much compatible to that for a physically acceptable stellar model. Arising solutions are also compared with graphical representations that provide strong evidences for more realistic and viable models, both at theoretical and astrophysical scale.

Variational Recurrent Neural Machine Translation ; Partially inspired by successful applications of variational recurrent neural networks, we propose a novel variational recurrent neural machine translation VRNMT model in this paper. Different from the variational NMT, VRNMT introduces a series of latent random variables to model the translation procedure of a sentence in a generative way, instead of a single latent variable. Specifically, the latent random variables are included into the hidden states of the NMT decoder with elements from the variational autoencoder. In this way, these variables are recurrently generated, which enables them to further capture strong and complex dependencies among the output translations at different timesteps. In order to deal with the challenges in performing efficient posterior inference and largescale training during the incorporation of latent variables, we build a neural posterior approximator, and equip it with a reparameterization technique to estimate the variational lower bound. Experiments on ChineseEnglish and EnglishGerman translation tasks demonstrate that the proposed model achieves significant improvements over both the conventional and variational NMT models.

Bouncing Unitary Cosmology II MiniSuperspace Phenomenology ; A companion paper provides a proposal for cosmic singularity resolution based upon general features of a bouncing unitary cosmological model in the minisuperspace approximation. This paper analyses novel phenomenology that can be identified within particular solutions of that model. First, we justify our choice of particular solutions based upon a clearly articulated and observationallymotivated principle. Second, we demonstrate that the chosen solutions follow a classical minisuperspace cosmology before smoothly bouncing off the classically singular region. Third, and most significantly, we identify a Rayleighscattering' limit for physically reasonable choices of parameters within which the solutions display effective behaviour that is insensitive to the details of rapidly oscillating Planckscale physics. This effective physics is found to be compatible with an effective period of cosmic inflation well below the Planck scale. The detailed effective physics of this Rayleighscattering limit is provided via i an exact analytical treatment of the model in the deSitter limit; and ii numerical solutions of the full model.

Computational model of avian nervous system nuclei governing learned song ; The means by which neuronal activity yields robust behavior is a ubiquitous question in neuroscience. In the songbird, the timing of a highly stereotyped song motif is attributed to the cortical nucleus HVC, and to feedback to HVC from downstream nuclei in the song motor pathway. Control of the acoustic structure appears to be shared by various structures, whose functional connectivity is largely unknown. Currently there exists no model for functional synaptic architecture that links HVC to song output in a manner consistent with experiments. Here we build on a previous model of HVC in which a distinct functional architecture may act as a pattern generator to drive downstream regions. Using a specific functional connectivity of the song motor pathway, we show how this HVC mechanism can generate simple representations of the driving forces for song. The model reproduces observed correlations between neuronal and respiratory activity and acoustic features of song. It makes testable predictions regarding the electrophysiology of distinct populations in the robust nucleus of the arcopallium RA, the connectivity within HVC and RA and between them, and the activity patterns of vocalrespiratory neurons in the brainstem.

Extension de la regression lineaire generalisee sur composantes supervisees SCGLR aux donnees groupees ; We address componentbased regularisation of a multivariate Generalized Linear Mixed Model. A set of random responses Y is modelled by a GLMM, using a set X of explanatory variables and a set T of additional covariates. Variables in X are assumed many and redundant generalized linear mixed regression demands regularisation with respect to X. By contrast, variables in T are assumed few and selected so as to demand no regularisation. Regularisation is performed building an appropriate number of orthogonal components that both contribute to model Y and capture relevant structural information in X. We propose to optimize a SCGLRspecific criterion within a Schall's algorithm in order to estimate the model. This extension of SCGLR is tested on simulated and real data, and compared to Ridgeand Lassobased regularisations.

On the Cosmological Frame Problem ; We introduce a fullyframe covariant formalism for inflation by taking into account conformal transformations in addition to field reparametrizations. We begin by providing a brief overview of frame problems in the history of science before outlining the crux of the frame problem in inflationary cosmology. After introducing the concept of frame tensors in curved field space, we demonstrate how the quantum perturbations and the observables sourced by them can be made frame covariant. We then specialize to twofield models, examining the impact of isocurvature effects on the inflationary observables in a framecovariant manner. We study the phenomenology of two particular models, a minimal polynomial model and a nonminimal model inspired by Higgs inflation. We observe that in the latter scenario, isocurvature effects are greatly enhanced. Moving beyond the treelevel approximation, we outline how our approach may be extended at the quantum level through the VilkoviskyDe Witt formalism and the generalization of frame tensors to configuration space, leading to a fully frameinvariant effective action. Finally, we summarize our findings and present possible future directions of research on the topic of frame covariance.

DiscrimNet SemiSupervised Action Recognition from Videos using Generative Adversarial Networks ; We propose an action recognition framework using Gen erative Adversarial Networks. Our model involves train ing a deep convolutional generative adversarial network DCGAN using a large video activity dataset without la bel information. Then we use the trained discriminator from the GAN model as an unsupervised pretraining step and finetune the trained discriminator model on a labeled dataset to recognize human activities. We determine good network architectural and hyperparameter settings for us ing the discriminator from DCGAN as a trained model to learn useful representations for action recognition. Our semisupervised framework using only appearance infor mation achieves superior or comparable performance to the current stateoftheart semisupervised action recog nition methods on two challenging video activity datasets UCF101 and HMDB51.

Chiral Gravitational Waves and Baryon Superfluid Dark Matter ; We develop a unified model of darkgenesis and baryogenesis involving strongly interacting dark quarks, utilizing the gravitational anomaly of chiral gauge theories. In these models, both the visible and dark baryon asymmetries are generated by the gravitational anomaly induced by the presence of chiral primordial gravitational waves. We provide a concrete model of an SU2 gauge theory with two massless quarks. In this model, the dark quarks condense and form a dark baryon charge superfluid DBS, in which the Higgsmode acts as cold dark matter. We elucidate the essential features of this dark matter scenario and discuss its phenomenological prospects.

Social Influence with Recurrent Mobility with multiple options ; In this paper, we discuss the possible generalizations of the Social Influence with Recurrent Mobility SIRM model developed in Phys. Rev. Lett. 112, 158701 2014. Although the SIRM model worked approximately satisfying when US election was modelled, it has its limits it has been developed only for twoparty systems and can lead to unphysical behaviour when one of the parties has extreme vote share close to 0 or 1. We propose here generalizations to the SIRM model by its extension for multiparty systems that are mathematically wellposed in case of extreme vote shares, too, by handling the noise term in a different way. In addition, we show that our method opens new applications for the study of elections by using a new calibration procedure, and makes possible to analyse the influence of the free will creating a new party and other local effects for different commuting network topologies.

Crowd Behavior Simulation with Emotional Contagion in Unexpected Multihazard Situations ; In this paper we present a novel crowd simulation method by modeling the generation and contagion of panic emotion under multihazard circumstances. Specifically, we first classify hazards into different types transient and persistent, concurrent and nonconcurrent, static and dynamic based on their inherent characteristics. Then, we introduce the concept of perilous field for each hazard and further transform the critical level of the field to its invokedpanic emotion. After that, we propose an emotional contagion model to simulate the evolving process of panic emotion caused by multiple hazards in these situations. Finally, we introduce an Emotional Reciprocal Velocity Obstacles ERVO model to simulate the crowd behaviors by augmenting the traditional RVO model with emotional contagion, which combines the emotional impact and local avoidance together for the first time. Our experimental results show that this method can soundly generate realistic group behaviors as well as panic emotion dynamics in a crowd in multihazard environments.

Neutrino masses in a conformal multiHiggsdoublet model ; We construct a conformal version of a general multiHiggsdoublet model with additional righthanded neutrino gaugesinglets. Assuming a minimal extension of the scalar sector by a real singlet field, we show that the resulting model achieves the same attractive properties as the nonconformal theory, combining the seesaw mechanism and higherorder mass production to generate naturally light neutrino masses. Starting with dimensionless couplings only, all masses and energy scales in the theory including the heavy Majorana masses and the electroweak scale are obtained from dimensional transmutation via the ColemanWeinberg mechanism. A characteristic feature of the conformal model is the appearance of the scalon in the scalar spectrum.The mass of this particle, which can be expressed in terms of the masses of the other particles in the theory, is produced at the oneloop level. We establish a connection between the large seesaw scale and a suppression of the scalon interactions. The positivity condition for the squared scalon mass requires sufficiently large masses of the additional Higgs bosons balancing the contributions of the heavy neutrinos.

An Infinitesimal Probabilistic Model for Principal Component Analysis of Manifold Valued Data ; We provide a probabilistic and infinitesimal view of how the principal component analysis procedure PCA can be generalized to analysis of nonlinear manifold valued data. Starting with the probabilistic PCA interpretation of the Euclidean PCA procedure, we show how PCA can be generalized to manifolds in an intrinsic way that does not resort to linearization of the data space. The underlying probability model is constructed by mapping a Euclidean stochastic process to the manifold using stochastic development of Euclidean semimartingales. The construction uses a connection and bundles of covariant tensors to allow global transport of principal eigenvectors, and the model is thereby an example of how principal fiber bundles can be used to handle the lack of global coordinate system and orientations that characterizes manifold valued statistics. We show how curvature implies nonintegrability of the equivalent of Euclidean principal subspaces, and how the stochastic flows provide an alternative to explicit construction of such subspaces. We describe estimation procedures for inference of parameters and prediction of principal components, and we give examples of properties of the model on embedded surfaces.

NMR signals within the generalized Langevin model for fractional Brownian motion ; The methods of Nuclear Magnetic Resonance belong to the best developed and often used tools for studying random motion of particles in different systems, including soft biological tissues. In the longtime limit the current mathematical description of the experiments allows proper interpretation of measurements of normal and anomalous diffusion. The shortertime dynamics is however correctly considered only in a few works that do not go beyond the standard memoryless Langevin description of the Brownian motion BM. In the present work, the attenuation function St for an ensemble of spinbearing particles in a magneticfield gradient, expressed in a form applicable for any kind of stationary stochastic dynamics of spins with or without a memory, is calculated in the frame of the model of fractional BM. The solution of the model for particles trapped in a harmonic potential is obtained in an exceedingly simple way and used for the calculation of St. In the limit of free particles coupled to a fractal heat bath, the results compare favorably with experiments acquired in human neuronal tissues. The effect of the trap is demonstrated by introducing a simple model for the generalized diffusion coefficient of the particle.

The theories of BaldwinShi hypergraphs and their atomic models ; We show that the quantifier elimination result for the ShelahSpencer almost sure theories of sparse random graphs Gn,nalpha given by Laskowski in 7 extends to their various analogues. The analogues will be obtained as theories of generic structures of certain classes of finite structures with a notion of strong substructure induced by rank functions and we will call the generics BaldwinShi hypergraphs. In the process we give a method of constructing extensions whose relative rank' is negative but arbitrarily small in context. We give a necessary and sufficient condition for the theory of a BaldwinShi hypergraph to have atomic models. We further show that for certain well behaved classes of theories of BaldwinShi hypergraphs, the existentially closed models and the atomic models correspond.

Pathological Analysis of Stress Urinary Incontinence in Females using Artificial Neural Networks ; Objectives To mathematically investigate urethral pressure and influencing parameters of stress urinary incontinence SUI in women, with focus on the clinical aspects of the mathematical modeling. Method Several patients' data are extracted from UPP and urodynamic documents and their relation and affinities are modeled using an artificial neural network ANN model. The studied parameter is urethral pressure as a function of two variables the age of the patient and the position in which the pressure was measured across the urethra normalized length. Results The ANNgenerated surface, showing the relation between the chosen parameters and the urethral pressure in the studied patients, is more efficient than the surface generated by conventional mathematical methods for clinical analysis, with multisample analysis being obtained. For example, in elderly people, there are many lowpressure zones throughout the urethra length, indicating that there is more incontinence in old age. Conclusion The predictions of urethral pressure made by the trained neural network model in relation to the studied effective parameters can be used to build a medical assistance system in order to help clinicians diagnose urinary incontinence problems more efficiently.

On ScottBlair model with timevarying viscosity in linear viscoelasticity ; In a recent paper, Zhou et al. studied the timedependent properties of Glass Fiber Reinforced Polymers GFRP composites by using a new rheological model with a timevariable viscosity coefficient. This rheology is essentially based on a generalized ScottBlair model with timevarying viscosity coefficient involving RiemannLiouville fractional derivatives. Motivated by this study, in this note we suggest a different generalization of the ScottBlair model based on the application of Caputo fractional derivatives of a function with respect to another function. This new mathematical approach can be useful in viscoelasticity and diffusion processes in order to consider timedependent coefficients. We are able to find the exact analytic solution of the creep experiment based on our new approach and we can compare it with the results obtained by Zhou et al.

Computational methods in cardiovascular mechanics ; The introduction of computational models in cardiovascular sciences has been progressively bringing new and unique tools for the investigation of the physiopathology. Together with the dramatic improvement of imaging and measuring devices on one side, and of computational architectures on the other one, mathematical and numerical models have provided a new, clearly noninvasive, approach for understanding not only basic mechanisms but also patientspecific conditions, and for supporting the design and the development of new therapeutic options. The terminology in silico is, nowadays, commonly accepted for indicating this new source of knowledge added to traditional in vitro and in vivo investigations. The advantages of in silico methodologies are basically the low cost in terms of infrastructures and facilities, the reduced invasiveness and, in general, the intrinsic predictive capabilities based on the use of mathematical models. The disadvantages are generally identified in the distance between the real cases and their virtual counterpart required by the conceptual modeling that can be detrimental for the reliability of numerical simulations.

Deep Multiple Instance Learning for Zeroshot Image Tagging ; Inline with the success of deep learning on traditional recognition problem, several endtoend deep models for zeroshot recognition have been proposed in the literature. These models are successful to predict a single unseen label given an input image, but does not scale to cases where multiple unseen objects are present. In this paper, we model this problem within the framework of Multiple Instance Learning MIL. To the best of our knowledge, we propose the first endtoend trainable deep MIL framework for the multilabel zeroshot tagging problem. Due to its novel design, the proposed framework has several interesting features 1 Unlike previous deep MIL models, it does not use any offline procedure e.g., Selective Search or EdgeBoxes for bag generation. 2 During test time, it can process any number of unseen labels given their semantic embedding vectors. 3 Using only seen labels per image as weak annotation, it can produce a bounding box for each predicted labels. We experiment with the NUSWIDE dataset and achieve superior performance across conventional, zeroshot and generalized zeroshot tagging tasks.

Adversarial Generalized Method of Moments ; We provide an approach for learning deep neural net representations of models described via conditional moment restrictions. Conditional moment restrictions are widely used, as they are the language by which social scientists describe the assumptions they make to enable causal inference. We formulate the problem of estimating the underling model as a zerosum game between a modeler and an adversary and apply adversarial training. Our approach is similar in nature to Generative Adversarial Networks GAN, though here the modeler is learning a representation of a function that satisfies a continuum of moment conditions and the adversary is identifying violating moments. We outline ways of constructing effective adversaries in practice, including kernels centered by kmeans clustering, and random forests. We examine the practical performance of our approach in the setting of nonparametric instrumental variable regression.

Normal hierarchy neutrino mass model revisited with leptogenesis ; We have studied the scenario of baryogenesis via leptogenesis in an A4 flavor symmetric framework considering type I seesaw as the origin of neutrino mass. Because of the presence of the fifth generation right handed neutrino the model naturally generates nonzero reactor mixing angle. We have considered two vev alignments for the extra flavon eta and studied the consequences in detail. As a whole the additional flavon along with the extra right handed neutrinos allow us to study thermal leptogenesis by the decay of the lightest right handed neutrino present in the model. We have computed the matterantimatter asymmetry for both flavor dependent and flavor independent leptogenesis by considering a considerably wider range of right handed neutrino mass. Finally, we correlate the baryon asymmetry of the universe BAU with the model parameters and light neutrino masses.

Graphite Iterative Generative Modeling of Graphs ; Graphs are a fundamental abstraction for modeling relational data. However, graphs are discrete and combinatorial in nature, and learning representations suitable for machine learning tasks poses statistical and computational challenges. In this work, we propose Graphite, an algorithmic framework for unsupervised learning of representations over nodes in large graphs using deep latent variable generative models. Our model parameterizes variational autoencoders VAE with graph neural networks, and uses a novel iterative graph refinement strategy inspired by lowrank approximations for decoding. On a wide variety of synthetic and benchmark datasets, Graphite outperforms competing approaches for the tasks of density estimation, link prediction, and node classification. Finally, we derive a theoretical connection between message passing in graph neural networks and meanfield variational inference.

Bianchi typeI dark energy cosmology with powerlaw relation in BransDicke theory of gravitation ; In this paper, we have studied the interacting and noninteracting dark energy and dark matter in the spatially homogenous and anisotropic Bianchi typeI model in the Brans Dicke theory of gravitation. The field equations have been solved by using i powerlaw relation and ii by assuming scale factor in terms of redshift. Here we have considered two cases of an interacting and noninteracting dark energy scenario and obtained general results. It has been found that for suitable choice of interaction between dark energy and dark matter we can avoid the coincidence problem which appears in the model. Some physical aspects and stability of the models are discussed in detail. The statefinder diagnostic pair i.e. r, s is adopted to differentiate our dark energy models.

Analysis of Rp inflationary model as pgeqslant 2 ; We study the Rp inflationary model of Muller1989rp for p2 using the result of Ref. Motohashi2014tra. After reproducing the observable quantities the power spectral index ns, its corresponding running alphafracdnsdlnk and the tensor to scalar ration r in terms of efolding number N and p, we show that Rp inflation model is still alive as p is from 2 to 2.02. In this range, our calculation confirms that ns and r agree with observations and alpha is of order 104 which needs more precise observational constraints. We find that, as the value of p increases, all ns, r and alpha decrease. However, the precise interdependence between these observables is such that this class of models can in principle be tested by the next generation of dedicated satellite CMB probes.

Video Prediction with Appearance and Motion Conditions ; Video prediction aims to generate realistic future frames by learning dynamic visual patterns. One fundamental challenge is to deal with future uncertainty How should a model behave when there are multiple correct, equally probable future We propose an AppearanceMotion Conditional GAN to address this challenge. We provide appearance and motion information as conditions that specify how the future may look like, reducing the level of uncertainty. Our model consists of a generator, two discriminators taking charge of appearance and motion pathways, and a perceptual ranking module that encourages videos of similar conditions to look similar. To train our model, we develop a novel conditioning scheme that consists of different combinations of appearance and motion conditions. We evaluate our model using facial expression and human action datasets and report favorable results compared to existing methods.

Evolution of Cooperation on Stochastic Block Models ; Cooperation is a major factor in the evolution of human societies. The structure of human social networks, which affects the dynamics of cooperation and other interpersonal phenomena, have common structural signatures. One of these signatures is the tendency to organize as groups. Among the generative models that network theorists use to emulate this feature is the Stochastic Block Model SBM. In this paper, we study evolutionary game dynamics on SBM networks. Using a recentlydiscovered duality between evolutionary games and coalescing random walks, we obtain analytical conditions such that natural selection favors cooperation over defection. We calculate the transition point for each community to favor cooperation. We find that a critical intercommunity link creation probability exists for given group density, such that the overall network supports cooperation even if individual communities inhibit it. As a byproduct, we present meanfield solutions for the critical benefittocost ratio which perform with remarkable accuracy for diverse generative network models, including those with community structure and heavytailed degree distributions. We also demonstrate the generalizability of the results to arbitrary twoplayer games.

Structured Synthesis for Probabilistic Systems ; We introduce the concept of structured synthesis for Markov decision processes where the structure is induced from finitely many prespecified options for a system configuration. The resulting synthesis problem is in general a nonlinear programming problem NLP with integer variables. As solving NLPs is in general not feasible, we present an alternative approach. We present a transformation of models specified in the PRISM probabilistic programming language to models that account for all possible system configurations by means of nondeterministic choices. Together with a control module that ensures consistent configurations throughout the system, this transformation enables the use of optimized tools for model checking in a blackbox fashion. While this transformation increases the size of a model, experiments with standard benchmarks show that the method provides a feasible approach for structured synthesis. Moreover, we demonstrate the usefulness along a realistic case study involving surveillance by unmanned aerial vehicles in a shipping facility.

Renormalization Scheme Ambiguities in the Models with More than One Coupling ; The process of renormalization to eliminate divergences arising in quantum field theory is not uniquely defined; one can always perform a finite renormalization, rendering finite perturbative results ambiguous. The consequences of making such finite renormalizations have been examined in the case of there being one or two couplings. In this paper we consider how finite renormalizations can affect more general models in which there are more than two couplings. In particular, we consider the Standard Model in which there are essentially five couplings. We show that in this model when neglecting all mass parameters if we use mass independent renormalization, then the renormalization group betafunctions are not unique beyond one loop order, that it is not in general possible to eliminate all terms beyond certain order for all these betafunctions, but that for a physical process all contributions beyond one loop order can be subsumed into the betafunctions.

Model Agnostic Saliency for Weakly Supervised Lesion Detection from Breast DCEMRI ; There is a heated debate on how to interpret the decisions provided by deep learning models DLM, where the main approaches rely on the visualization of salient regions to interpret the DLM classification process. However, these approaches generally fail to satisfy three conditions for the problem of lesion detection from medical images 1 for images with lesions, all salient regions should represent lesions, 2 for images containing no lesions, no salient region should be produced,and 3 lesions are generally small with relatively smooth borders. We propose a new modelagnostic paradigm to interpret DLM classification decisions supported by a novel definition of saliency that incorporates the conditions above. Our modelagnostic 1class saliency detector MASD is tested on weakly supervised breast lesion detection from DCEMRI, achieving stateoftheart detection accuracy when compared to current visualization methods.

Score Permutation Based Finite Sample Inference for Generalized AutoRegressive Conditional Heteroskedasticity GARCH Models ; A standard model of conditional heteroscedasticity, i.e., the phenomenon that the variance of a process changes over time, is the Generalized AutoRegressive Conditional Heteroskedasticity GARCH model, which is especially important for economics and finance. GARCH models are typically estimated by the QuasiMaximum Likelihood QML method, which works under mild statistical assumptions. Here, we suggest a finite sample approach, called ScoPe, to construct distributionfree confidence regions around the QML estimate, which have exact coverage probabilities, despite no additional assumptions about moments are made. ScoPe is inspired by the recently developed SignPerturbed Sums SPS method, which however cannot be applied in the GARCH case. ScoPe works by perturbing the score function using randomly permuted residuals. This produces alternative samples which lead to exact confidence regions. Experiments on simulated and stock market data are also presented, and ScoPe is compared with the asymptotic theory and bootstrap approaches.

Ghostfree higher order gravity from bigravity ; In this work we studied the higher order gravity model which corresponds to HassanRosen ghostfree bigravity. To do this we absorb one of the metrics in bigravity model in favor of the other metric in a recursive way. For the second recursion step we get conformal gravity same as what has been done in the literature. We generalize this idea by calculating up to the fourth order gravity. To reach to a ghostfree higher order gravity we need to go to infinitederivative order gravity but we can see our model as an effective theory. So adding higher order terms results in wider range of validity of our model. We emphasize that graviton mass controls the validity of perturbative approach.

Grounding Visual Explanations ; Existing visual explanation generating agents learn to fluently justify a class prediction. However, they may mention visual attributes which reflect a strong class prior, although the evidence may not actually be in the image. This is particularly concerning as ultimately such agents fail in building trust with human users. To overcome this limitation, we propose a phrasecritic model to refine generated candidate explanations augmented with flipped phrases which we use as negative examples while training. At inference time, our phrasecritic model takes an image and a candidate explanation as input and outputs a score indicating how well the candidate explanation is grounded in the image. Our explainable AI agent is capable of providing counter arguments for an alternative prediction, i.e. counterfactuals, along with explanations that justify the correct classification decisions. Our model improves the textual explanation quality of finegrained classification decisions on the CUB dataset by mentioning phrases that are grounded in the image. Moreover, on the FOIL tasks, our agent detects when there is a mistake in the sentence, grounds the incorrect phrase and corrects it significantly better than other models.

Analysing dissipative effects in the CDM model ; In the present paper, the effects of viscous dark matter are analysed within the LambdaCDM model. Here we consider bulk viscosity through the IsraelStewart theory approach, leading to an effective pressure term in the continuity equation that accomplishes for the dissipative effects of the dark matter fluid. Then, the corresponding equation for viscosity is solved and a general equation for the Hubble parameter is obtained with the presence of a cosmological constant. The existence of de Sitter solutions is discussed, where a wider range of solutions is found in comparison to the LambdaCDM model. Also the conditions for the near thermodynamical equilibrium of the fluid is analysed. Finally, a qualitative analysis provides some constraints on the model by using Supernovae Ia data which reveals the possible importance of causal thermodynamics in cosmology.

Scaling and bias codes for modeling speakeradaptive DNNbased speech synthesis systems ; Most neuralnetwork based speakeradaptive acoustic models for speech synthesis can be categorized into either layerbased or inputcode approaches. Although both approaches have their own pros and cons, most existing works on speaker adaptation focus on improving one or the other. In this paper, after we first systematically overview the common principles of neuralnetwork based speakeradaptive models, we show that these approaches can be represented in a unified framework and can be generalized further. More specifically, we introduce the use of scaling and bias codes as generalized means for speakeradaptive transformation. By utilizing these codes, we can create a more efficient factorized speakeradaptive model and capture advantages of both approaches while reducing their disadvantages. The experiments show that the proposed method can improve the performance of speaker adaptation compared with speaker adaptation based on the conventional input code.

Attention is All We Need Nailing Down Objectcentric Attention for Egocentric Activity Recognition ; In this paper we propose an endtoend trainable deep neural network model for egocentric activity recognition. Our model is built on the observation that egocentric activities are highly characterized by the objects and their locations in the video. Based on this, we develop a spatial attention mechanism that enables the network to attend to regions containing objects that are correlated with the activity under consideration. We learn highly specialized attention maps for each frame using classspecific activations from a CNN pretrained for generic image recognition, and use them for spatiotemporal encoding of the video with a convolutional LSTM. Our model is trained in a weakly supervised setting using raw videolevel activityclass labels. Nonetheless, on standard egocentric activity benchmarks our model surpasses by up to 6 points recognition accuracy the currently best performing method that leverages hand segmentation and object location strong supervision for training. We visually analyze attention maps generated by the network, revealing that the network successfully identifies the relevant objects present in the video frames which may explain the strong recognition performance. We also discuss an extensive ablation analysis regarding the design choices.

Learning to Transfer Unsupervised Meta Domain Translation ; Unsupervised domain translation has recently achieved impressive performance with Generative Adversarial Network GAN and sufficient unpaired training data. However, existing domain translation frameworks form in a disposable way where the learning experiences are ignored and the obtained model cannot be adapted to a new coming domain. In this work, we take on unsupervised domain translation problems from a metalearning perspective. We propose a model called MetaTranslation GAN MTGAN to find good initialization of translation models. In the metatraining procedure, MTGAN is explicitly trained with a primary translation task and a synthesized dual translation task. A cycleconsistency metaoptimization objective is designed to ensure the generalization ability. We demonstrate effectiveness of our model on ten diverse twodomain translation tasks and multiple face identity translation tasks. We show that our proposed approach significantly outperforms the existing domain translation methods when each domain contains no more than ten training samples.

Robust SequencetoSequence Acoustic Modeling with Stepwise Monotonic Attention for Neural TTS ; Neural TTS has demonstrated strong capabilities to generate humanlike speech with high quality and naturalness, while its generalization to outofdomain texts is still a challenging task, with regard to the design of attentionbased sequencetosequence acoustic modeling. Various errors occur in those inputs with unseen context, including attention collapse, skipping, repeating, etc., which limits the broader applications. In this paper, we propose a novel stepwise monotonic attention method in sequencetosequence acoustic modeling to improve the robustness on outofdomain inputs. The method utilizes the strict monotonic property in TTS with constraints on monotonic hard attention that the alignments between inputs and outputs sequence must be not only monotonic but allowing no skipping on inputs. Soft attention could be used to evade mismatch between training and inference. The experimental results show that the proposed method could achieve significant improvements in robustness on outofdomain scenarios for phonemebased models, without any regression on the indomain naturalness test.

NeuralVis Visualizing and Interpreting Deep Learning Models ; Deep Neural NetworkDNN techniques have been prevalent in software engineering. They are employed to faciliatate various software engineering tasks and embedded into many software applications. However, analyzing and understanding their behaviors is a difficult task for software engineers. In this paper, to support software engineers in visualizing and interpreting deep learning models, we present NeuralVis, an instancebased visualization tool for DNN. NeuralVis is designed for 1. visualizing the structure of DNN models, i.e., components, layers, as well as connections; 2. visualizing the data transformation process; 3. integrating existing adversarial attack algorithms for test input generation; 4. comparing intermediate outputs of different instances to guide the test input generation; To demonstrate the effectiveness of NeuralVis, we conduct an user study involving ten participants on two classic DNN models, i.e., LeNet and VGG12. The result shows NeuralVis can assist developers in identifying the critical features that determines the prediction results. Video httpsyoutu.behRxCovrOZFI

KERMIT Generative InsertionBased Modeling for Sequences ; We present KERMIT, a simple insertionbased approach to generative modeling for sequences and sequence pairs. KERMIT models the joint distribution and its decompositions i.e., marginals and conditionals using a single neural network and, unlike much prior work, does not rely on a prespecified factorization of the data distribution. During training, one can feed KERMIT paired data x, y to learn the joint distribution px, y, and optionally mix in unpaired data x or y to refine the marginals px or py. During inference, we have access to the conditionals px mid y and py mid x in both directions. We can also sample from the joint distribution or the marginals. The model supports both serial fully autoregressive decoding and parallel partially autoregressive decoding, with the latter exhibiting an empirically logarithmic runtime. We demonstrate through experiments in machine translation, representation learning, and zeroshot cloze question answering that our unified approach is capable of matching or exceeding the performance of dedicated stateoftheart systems across a wide range of tasks without the need for problemspecific architectural adaptation.

Bad Global Minima Exist and SGD Can Reach Them ; Several works have aimed to explain why overparameterized neural networks generalize well when trained by Stochastic Gradient Descent SGD. The consensus explanation that has emerged credits the randomized nature of SGD for the bias of the training process towards lowcomplexity models and, thus, for implicit regularization. We take a careful look at this explanation in the context of image classification with common deep neural network architectures. We find that if we do not regularize emphexplicitly, then SGD can be easily made to converge to poorlygeneralizing, highcomplexity models all it takes is to first train on a random labeling on the data, before switching to properly training with the correct labels. In contrast, we find that in the presence of explicit regularization, pretraining with random labels has no detrimental effect on SGD. We believe that our results give evidence that explicit regularization plays a far more important role in the success of overparameterized neural networks than what has been understood until now. Specifically, by penalizing complicated models independently of their fit to the data, regularization affects training dynamics also far away from optima, making simple models that fit the data well discoverable by local methods, such as SGD.

Conversing by Reading Contentful Neural Conversation with Ondemand Machine Reading ; Although neural conversation models are effective in learning how to produce fluent responses, their primary challenge lies in knowing what to say to make the conversation contentful and nonvacuous. We present a new endtoend approach to contentful neural conversation that jointly models response generation and ondemand machine reading. The key idea is to provide the conversation model with relevant longform text on the fly as a source of external knowledge. The model performs QAstyle reading comprehension on this text in response to each conversational turn, thereby allowing for more focused integration of external knowledge than has been possible in prior approaches. To support further research on knowledgegrounded conversation, we introduce a new largescale conversation dataset grounded in external web pages 2.8M turns, 7.4M sentences of grounding. Both human evaluation and automated metrics show that our approach results in more contentful responses compared to a variety of previous methods, improving both the informativeness and diversity of generated output.

Importance Weighted Adversarial Variational Autoencoders for Spike Inference from Calcium Imaging Data ; The Importance Weighted Auto Encoder IWAE objective has been shown to improve the training of generative models over the standard Variational Auto Encoder VAE objective. Here, we derive importance weighted extensions to AVB and AAE. These latent variable models use implicitly defined inference networks whose approximate posterior density qphizx cannot be directly evaluated, an essential ingredient for importance weighting. We show improved training and inference in latent variable models with our adversarially trained importance weighting method, and derive new theoretical connections between adversarial generative model training criteria and marginal likelihood based methods. We apply these methods to the important problem of inferring spiking neural activity from calcium imaging data, a challenging posterior inference problem in neuroscience, and show that posterior samples from the adversarial methods outperform factorized posteriors used in VAEs.

Selection consistency of Lassobased procedures for misspecified highdimensional binary model and random regressors ; We consider selection of random predictors for highdimensional regression problem with binary response for a general loss function. Important special case is when the binary model is semiparametric and the response function is misspecified under parametric model fit. Selection for such a scenario aims at recovering the support of the minimizer of the associated risk with large probability. We propose a twostep selection procedure which consists of screening and ordering predictors by Lasso method and then selecting a subset of predictors which minimizes Generalized Information Criterion on the corresponding nested family of models. We prove consistency of the selection method under conditions which allow for much larger number of predictors than number of observations. For the semiparametric case when distribution of random predictors satisfies linear regression conditions the true and the estimated parameters are collinear and their common support can be consistently identified.

A ModelBased General Alternative to the Standardised Precipitation Index ; In this paper, we introduce two new modelbased versions of the widelyused standardized precipitation index SPI for detecting and quantifying the magnitude of extreme hydroclimatic events. Our analytical approach is based on generalized additive models for location, scale and shape GAMLSS, which helps as to overcome some limitations of the SPI. We compare our modelbased standardised indices MBSIs with the SPI using precipitation data collected between January 2004 December 2013 522 weeks in Caapiranga, a roadless municipality of Amazonas State. As a result, it is shown that the MBSI1 is an index with similar properties to the SPI, but with improved methodology. In comparison to the SPI, our MBSI1 index allows for the use of different zeroaugmented distributions, it works with more flexible timescales, can be applied to shorter records of data and also takes into account temporal dependencies in known seasonal behaviours. Our approach is implemented in an R package, mbsi, available from Github.

XLNet Generalized Autoregressive Pretraining for Language Understanding ; With the capability of modeling bidirectional contexts, denoising autoencoding based pretraining like BERT achieves better performance than pretraining approaches based on autoregressive language modeling. However, relying on corrupting the input with masks, BERT neglects dependency between the masked positions and suffers from a pretrainfinetune discrepancy. In light of these pros and cons, we propose XLNet, a generalized autoregressive pretraining method that 1 enables learning bidirectional contexts by maximizing the expected likelihood over all permutations of the factorization order and 2 overcomes the limitations of BERT thanks to its autoregressive formulation. Furthermore, XLNet integrates ideas from TransformerXL, the stateoftheart autoregressive model, into pretraining. Empirically, under comparable experiment settings, XLNet outperforms BERT on 20 tasks, often by a large margin, including question answering, natural language inference, sentiment analysis, and document ranking.

A New Framework for software Library Investment Metrics ; Software quality is considered as one of the most important challenges in software engineering. It has many dimensions which differ from users' point of view that depend on their requirements. Therefore, those dimensions lead to difficulty in measuring and defining the software quality properly. Software quality measurement is the main core of the software quality. Thus, it is necessary to study and develop the software measurements to meet the better quality. The use of libraries increases software quality more than that of using generic programming because these libraries are prepared and tested in advance. In addition, these libraries reduce the effort that is spent in designing, testing, and maintenance processes. In this research, we presented a new model to calculate the saved effort that results from using libraries instead of generic programming in the coding, testing, and productivity processes. The proposed model consists of three metrics that are Library Investment Ratio, Library Investment Level, and Program Simplicity. An empirical analyzes has been applied into ten projects to compare the results of the model with Reuse Percent. The results show that the model has better indication of the improvement of software quality and productivity rather than Reuse Percent.

Alchemy A Quantum Chemistry Dataset for Benchmarking AI Models ; We introduce a new molecular dataset, named Alchemy, for developing machine learning models useful in chemistry and material science. As of June 20th 2019, the dataset comprises of 12 quantum mechanical properties of 119,487 organic molecules with up to 14 heavy atoms, sampled from the GDB MedChem database. The Alchemy dataset expands the volume and diversity of existing molecular datasets. Our extensive benchmarks of the stateoftheart graph neural network models on Alchemy clearly manifest the usefulness of new data in validating and developing machine learning models for chemistry and material science. We further launch a contest to attract attentions from researchers in the related fields. More details can be found on the contest website footnotehttpsalchemy.tencent.com. At the time of benchamrking experiment, we have generated 119,487 molecules in our Alchemy dataset. More molecular samples are generated since then. Hence, we provide a list of molecules used in the reported benchmarks.

Learning partial correlation graphs and graphical models by covariance queries ; We study the problem of recovering the structure underlying large Gaussian graphical models or, more generally, partial correlation graphs. In highdimensional problems it is often too costly to store the entire sample covariance matrix. We propose a new input model in which one can query single entries of the covariance matrix. We prove that it is possible to recover the support of the inverse covariance matrix with low query and computational complexity. Our algorithms work in a regime when this support is represented by treelike graphs and, more generally, for graphs of small treewidth. Our results demonstrate that for large classes of graphs, the structure of the corresponding partial correlation graphs can be determined much faster than even computing the empirical covariance matrix.

Causal Regularization ; I argue that regularizing terms in standard regression methods not only help against overfitting finite data, but sometimes also yield better causal models in the infinite sample regime. I first consider a multidimensional variable linearly influencing a target variable with some multidimensional unobserved common cause, where the confounding effect can be decreased by keeping the penalizing term in Ridge and Lasso regression even in the population limit. Choosing the size of the penalizing term, is however challenging, because cross validation is pointless. Here it is done by first estimating the strength of confounding via a method proposed earlier, which yielded some reasonable results for simulated and real data. Further, I prove a causal generalization bound' which states subject to a particular model of confounding that the error made by interpreting any nonlinear regression as causal model can be bounded from above whenever functions are taken from a not too rich class. In other words, the bound guarantees generalization from observational to interventional distributions, which is usually not subject of statistical learning theory and is only possible due to the underlying symmetries of the confounder model.

FVA Modeling Perceived Friendliness of Virtual Agents Using Movement Characteristics ; We present a new approach for improving the friendliness and warmth of a virtual agent in an AR environment by generating appropriate movement characteristics. Our algorithm is based on a novel datadriven friendliness model that is computed using a userstudy and psychological characteristics. We use our model to control the movements corresponding to the gaits, gestures, and gazing of friendly virtual agents FVAs as they interact with the user's avatar and other agents in the environment. We have integrated FVA agents with an AR environment using with a Microsoft HoloLens. Our algorithm can generate plausible movements at interactive rates to increase the social presence. We also investigate the perception of a user in an AR setting and observe that an FVA has a statistically significant improvement in terms of the perceived friendliness and social presence of a user compared to an agent without the friendliness modeling. We observe an increment of 5.71 in the mean responses to a friendliness measure and an improvement of 4.03 in the mean responses to a social presence measure.

Reducedorder surrogate models for scalartensor gravity in the strong field and applications to binary pulsars and GW170817 ; We investigate the scalartensor gravity of Damour and EspositoFarese DEF, which predicts nontrivial phenomena in the nonperturbative strongfield regime for neutron stars NSs. Instead of solving the modified TolmanOppenheimerVolkoff equations, we construct reducedorder surrogate models, coded in the pySTGROM package, to predict the relations of a NS radius, mass, and effective scalar coupling to its central density. Our models are accurate at sim1 level and speed up largescale calculations by two orders of magnitude. As an application, we use pySTGROM and Markovchain Monte Carlo techniques to constrain parameters in the DEF theory, with five welltimed binary pulsars, the binary NS BNS inspiral GW170817, and a hypothetical BNS inspiral in the Advanced LIGO and nextgeneration GW detectors. In the future, as more binary pulsars and BNS mergers are detected, our surrogate models will be helpful in constraining strongfield gravity with essential speed and accuracy.

Understanding Memory Modules on Learning Simple Algorithms ; Recent work has shown that memory modules are crucial for the generalization ability of neural networks on learning simple algorithms. However, we still have little understanding of the working mechanism of memory modules. To alleviate this problem, we apply a twostep analysis pipeline consisting of first inferring hypothesis about what strategy the model has learned according to visualization and then verify it by a novel proposed qualitative analysis method based on dimension reduction. Using this method, we have analyzed two popular memoryaugmented neural networks, neural Turing machine and stackaugmented neural network on two simple algorithm tasks including reversing a random sequence and evaluation of arithmetic expressions. Results have shown that on the former task both models can learn to generalize and on the latter task only the stackaugmented model can do so. We show that different strategies are learned by the models, in which specific categories of input are monitored and different policies are made based on that to change the memory.

Rare Disease Detection by Sequence Modeling with Generative Adversarial Networks ; Rare diseases affecting 350 million individuals are commonly associated with delay in diagnosis or misdiagnosis. To improve those patients' outcome, rare disease detection is an important task for identifying patients with rare conditions based on longitudinal medical claims. In this paper, we present a deep learning method for detecting patients with exocrine pancreatic insufficiency EPI a rare disease. The contribution includes 1 a large longitudinal study using 7 years medical claims from 1.8 million patients including 29,149 EPI patients, 2 a new deep learning model using generative adversarial networks GANs to boost rare disease class, and also leveraging recurrent neural networks to model patient sequence data, 3 an accurate prediction with 0.56 PRAUC which outperformed benchmark models in terms of precision and recall.

VELC A New Variational AutoEncoder Based Model for Time Series Anomaly Detection ; Anomaly detection is a classical but worthwhile problem, and many deep learningbased anomaly detection algorithms have been proposed, which can usually achieve better detection results than traditional methods. In view of reconstruct ability of the model and the calculation of anomaly score, this paper proposes a time series anomaly detection method based on Variational AutoEncoder modelVAE with reEncoder and Latent Constraint networkVELC. In order to modify reconstruct ability of the model to prevent it from reconstructing abnormal samples well, we add a constraint network in the latent space of the VAE to force it generate new latent variables that are similar with that of training samples. To be able to calculate anomaly score in two feature spaces, we train a reencoder to transform the generated data to a new latent space. For better handling the time series, we use the LSTM as the encoder and decoder part of the VAE framework. Experimental results of several benchmarks show that our method outperforms stateoftheart anomaly detection methods.

Quintessential inflation with a trap and axionic dark matter ; We study a new model of quintessential inflation which is inspired by supergravity and string theory. The model features a kinetic pole, which gives rise to the inflationary plateau, and a runaway quintessential tail. We envisage a coupling between the inflaton and the PecceiQuinn PQ field which terminates the roll of the runaway inflaton and traps the latter at an enhanced symmetry point ESP, thereby breaking the PQ symmetry. The kinetic density of the inflaton is transferred to the newly created thermal bath of the hot big bang due to the decay of PQ particles. The model successfully accounts for the observations of inflation and dark energy without any finetuning, while also resolving the strong CP problem of QCD and generating axionic dark matter, without isocurvature perturbations. Trapping the inflaton at the ESP ensures that the model does not suffer from the infamous 5th force problem, which typically plagues quintessence.

Propagation of Polar Gravitational Waves in fR,T Scenario ; This paper investigates the propagation of polar gravitational waves in the spatially flat FRW universe consisting of a perfect fluid in the scenario of R2lambda T model of fR,T gravity lambda being the model parameter. The spatially flat universe model is perturbed via ReggeWheeler perturbations inducing polar gravitational waves and the field equations are formulated for both unperturbed as well as perturbed spacetimes. We solve these field equations simultaneously for the perturbation parameters introduced in the metric, matter, and velocity in the radiation, as well as dark energy, dominated phases. It is found that the polar gravitational waves can produce changes in the background matter distribution as well as velocity components in the radiation era similar to general relativity case. Moreover, we have discussed the impact of model parameter on the amplitude of gravitational waves.

Axionelectron decoupling in nucleophobic axion models ; The strongest upper bounds on the axion mass come from astrophysical observations like the neutrino burst duration of SN1987A, which depends on the axion couplings to nucleons, or the whitedwarf cooling rates and redgiant evolution, which involve the axionelectron coupling. It has been recently argued that in variants of DFSZ models with generationdependent PecceiQuinn charges an approximate axionnucleon decoupling can occur, strongly relaxing the SN1987A bound. However, as in standard DFSZ models, the axion remains in general coupled to electrons, unless an ad hoc cancellation is engineered. Here we show that axionelectron decoupling can be implemented without extra tunings in DFSZlike models with three Higgs doublets. Remarkably, the numerical value of the quark mass ratio mumdsim 12 is crucial to open up this possibility.

Fairness and Diversity in the Recommendation and Ranking of Participatory Media Content ; Online participatory media platforms that enable onetomany communication among users, see a significant amount of user generated content and consequently face a problem of being able to recommend a subset of this content to its users. We address the problem of recommending and ranking this content such that different viewpoints about a topic get exposure in a fair and diverse manner. We build our model in the context of a voicebased participatory media platform running in rural central India, for lowincome and lessliterate communities, that plays audio messages in a ranked list to users over a phone call and allows them to contribute their own messages. In this paper, we describe our model and evaluate it using calllogs from the platform, to compare the fairness and diversity performance of our model with the manual editorial processes currently being followed. Our models are generic and can be adapted and applied to other participatory media platforms as well.

Style Transfer Applied to Face Liveness Detection with UserCentered Models ; This paper proposes a face antispoofing usercentered model FASUCM. The major difficulty, in this case, is obtaining fraudulent images from all users to train the models. To overcome this problem, the proposed method is divided in three main parts generation of new spoof images, based on style transfer and spoof image representation models; training of a Convolutional Neural Network CNN for liveness detection; evaluation of the live and spoof testing images for each subject. The generalization of the CNN to perform style transfer has shown promising qualitative results. Preliminary results have shown that the proposed method is capable of distinguishing between live and spoof images on the SiW database, with an average classification error rate of 0.22.

Multicomponent scalar fields and the complete factorization of its equations of motion ; In the paper by Bazeia D. et al., EPL, 119 2017 61002, the authors demonstrate the equivalence between the secondorder differential equation of motion and a family of firstorder differential equations of Bogomolnyi type for the cases of single real and complex scalar field theories with noncanonical dynamics. The goal of this paper is to demonstrate that this equivalence is also valid for a more general classes of real scalar field models. We start the paper by demonstrating the equivalence in a single real scalar model. The first goal is to generalize the equivalence presented in papers by Bazeia et al. to a single real scalar field model without a specific form for its Lagrangian. The second goal is to use the setup presented in the first demonstration to show that this equivalence can be achieved also in a real multicomponent scalar field model again without a specific form for its Lagrangian. The main goal of this paper is to show that this equivalence can be achieved in real scalar field scenarios that can be standard, or nonstandard, with single, or multicomponent, scalar fields.

Linear growth index of matter perturbations in Rastall gravity ; Rastall gravity theory shows notable features consistent with physical observations in comparison to the standard Einstein theory. Recently, there has been a debate about the equivalence of Rastall gravity and general relativity. Motivated by this open issue, in the present work, we attempt to shed some light on this debate by analyzing the evolution of the Rastall based cosmological model at the background as well as perturbation level. Employing the dynamical system techniques, we found that at late times, the dynamics of the model resembles the LambdaCDM model at the background level irrespective of the choice of Rastall's parameter. However, at the perturbation level, we found that the evolution of the growth index heavily depends on the Rastall's parameter and displays a significant deviation from the LambdaCDM model.

SemiSupervised Learning by Disentangling and SelfEnsembling Over Stochastic Latent Space ; The success of deep learning in medical imaging is mostly achieved at the cost of a large labeled data set. Semisupervised learning SSL provides a promising solution by leveraging the structure of unlabeled data to improve learning from a small set of labeled data. Selfensembling is a simple approach used in SSL to encourage consensus among ensemble predictions of unknown labels, improving generalization of the model by making it more insensitive to the latent space. Currently, such an ensemble is obtained by randomization such as dropout regularization and random data augmentation. In this work, we hypothesize from the generalization perspective that selfensembling can be improved by exploiting the stochasticity of a disentangled latent space. To this end, we present a stacked SSL model that utilizes unsupervised disentangled representation learning as the stochastic embedding for selfensembling. We evaluate the presented model for multilabel classification using chest Xray images, demonstrating its improved performance over related SSL models as well as the interpretability of its disentangled representations.

Meanfield solution of structural balance dynamics in nonzero temperature ; In signed networks with simultaneous friendly and hostile interactions, there is a general tendency to a global structural balance, based on the dynamical model of links status. Although the structural balance represents a state of the network with a lack of contentious situations, there are always tensions in real networks. To study such networks, we generalize the balance dynamics in nonzero temperatures. The presented model uses elements from BoltzmannGibbs statistical physics to assign an energy to each type of triad, and it introduces the temperature as a measure of tension tolerance of the network. Based on the meanfield solution of the model, we find out that the model undergoes a firstorder phase transition from an imbalanced random state to structural balance with a critical temperature Tc, where in the case of T Tc there is no chance to reach the balanced state. A main feature of the firstorder phase transition is the occurrence of a hysteresis loop crossing the balanced and imbalanced regimes.

Stochastic trajectory prediction with social graph network ; Pedestrian trajectory prediction is a challenging task because of the complexity of realworld human social behaviors and uncertainty of the future motion. For the first issue, existing methods adopt fully connected topology for modeling the social behaviors, while ignoring nonsymmetric pairwise relationships. To effectively capture social behaviors of relevant pedestrians, we utilize a directed social graph which is dynamically constructed on timely location and speed direction. Based on the social graph, we further propose a network to collect social effects and accumulate with individual representation, in order to generate destinationoriented and socialaware representations. For the second issue, instead of modeling the uncertainty of the entire future as a whole, we utilize a temporal stochastic method for sequentially learning a prior model of uncertainty during social interactions. The prediction on the next step is then generated by sampling on the prior model and progressively decoding with a hierarchical LSTMs. Experimental results on two public datasets show the effectiveness of our method, especially when predicting trajectories in very crowded scenes.

Joint Adversarial Training Incorporating both Spatial and Pixel Attacks ; Conventional adversarial training methods using attacks that manipulate the pixel value directly and individually, leading to models that are less robust in face of spatial transformationbased attacks. In this paper, we propose a joint adversarial training method that incorporates both spatial transformationbased and pixelvalue based attacks for improving model robustness. We introduce a spatial transformationbased attack with an explicit notion of budget and develop an algorithm for spatial attack generation. We further integrate both pixel and spatial attacks into one generation model and show how to leverage the complementary strengths of each other in training for improving the overall model robustness. Extensive experimental results on different benchmark datasets compared with stateoftheart methods verified the effectiveness of the proposed method.

Leveraging Pretrained Checkpoints for Sequence Generation Tasks ; Unsupervised pretraining of large neural models has recently revolutionized Natural Language Processing. By warmstarting from the publicly released checkpoints, NLP practitioners have pushed the stateoftheart on multiple benchmarks while saving significant amounts of compute time. So far the focus has been mainly on the Natural Language Understanding tasks. In this paper, we demonstrate the efficacy of pretrained checkpoints for Sequence Generation. We developed a Transformerbased sequencetosequence model that is compatible with publicly available pretrained BERT, GPT2 and RoBERTa checkpoints and conducted an extensive empirical study on the utility of initializing our model, both encoder and decoder, with these checkpoints. Our models result in new stateoftheart results on Machine Translation, Text Summarization, Sentence Splitting, and Sentence Fusion.

CVA and vulnerable options in stochastic volatility models ; In this work we want to provide a general principle to evaluate the CVA Credit Value Adjustment for a vulnerable option, that is an option subject to some default event, concerning the solvability of the issuer. CVA is needed to evaluate correctly the contract and it is particularly important in presence of WWR Wrong Way Risk, when a credit deterioration determines an increase of the claim's price. In particular, we are interested in evaluating the CVA in stochastic volatility models for the underlying's price which often fit quite well the market's prices when admitting correlation with the default event. By cunningly using Ito's calculus, we provide a general representation formula applicable to some popular models such as SABR, Hull White and Heston, which explicitly shows the correction in CVA due to the processes correlation. Later, we specialize this formula and construct its approximation for the three selected models. Lastly, we run a numerical study to test the formula's accuracy, comparing our results with Monte Carlo simulations.

Climatedriven statistical models as effective predictors of local dengue incidence in Costa Rica A Generalized Additive Model and Random Forest approach ; Climate has been an important factor in shaping the distribution and incidence of dengue cases in tropical and subtropical countries. In Costa Rica, a tropical country with distinctive microclimates, dengue has been endemic since its introduction in 1993, inflicting substantial economic, social, and public health repercussions. Using the number of dengue reported cases and climate data from 20072017, we fitted a prediction model applying a Generalized Additive Model GAM and Random Forest RF approach, which allowed us to retrospectively predict the relative risk of dengue in five climatological diverse municipalities around the country.

Towards Digital Retina in Smart Cities A Model Generation, Utilization and Communication Paradigm ; The digital retina in smart cities is to select what the City Eye tells the City Brain, and convert the acquired visual data from frontend visual sensors to features in an intelligent sensing manner. By deploying deep learning andor handcrafted models in frontend devices, the compact features can be extracted and subsequently delivered to backend cloud for search and advanced analytics. In this context, we propose a model generation, utilization, and communication paradigm, aiming to address a set of unique challenges for better artificial intelligence services in smart cities. In particular, we present an integrated multiple deep learning models reuse and prediction strategy, which greatly increases the feasibility of the digital retina in processing and analyzing the largescale visual data in smart cities. The promise of the proposed paradigm is demonstrated through a set of experiments.

An EndtoEnd Graph Convolutional Kernel Support Vector Machine ; A novel kernelbased support vector machine SVM for graph classification is proposed. The SVM feature space mapping consists of a sequence of graph convolutional layers, which generates a vector space representation for each vertex, followed by a pooling layer which generates a reproducing kernel Hilbert space RKHS representation for the graph. The use of a RKHS offers the ability to implicitly operate in this space using a kernel function without the computational complexity of explicitly mapping into it. The proposed model is trained in a supervised endtoend manner whereby the convolutional layers, the kernel function and SVM parameters are jointly optimized with respect to a regularized classification loss. This approach is distinct from existing kernelbased graph classification models which instead either use feature engineering or unsupervised learning to define the kernel function. Experimental results demonstrate that the proposed model outperforms existing deep learning baseline models on a number of datasets.

Finegrained VideoText Retrieval with Hierarchical Graph Reasoning ; Crossmodal retrieval between videos and texts has attracted growing attentions due to the rapid emergence of videos on the web. The current dominant approach for this problem is to learn a joint embedding space to measure crossmodal similarities. However, simple joint embeddings are insufficient to represent complicated visual and textual details, such as scenes, objects, actions and their compositions. To improve finegrained videotext retrieval, we propose a Hierarchical Graph Reasoning HGR model, which decomposes videotext matching into globaltolocal levels. To be specific, the model disentangles texts into hierarchical semantic graph including three levels of events, actions, entities and relationships across levels. Attentionbased graph reasoning is utilized to generate hierarchical textual embeddings, which can guide the learning of diverse and hierarchical video representations. The HGR model aggregates matchings from different videotext levels to capture both global and local details. Experimental results on three videotext datasets demonstrate the advantages of our model. Such hierarchical decomposition also enables better generalization across datasets and improves the ability to distinguish finegrained semantic differences.