Search

Generating dances that are both lifelike and well-aligned with music continues to be a challenging task in the cross-modal domain. This paper introduces PopDanceSet, the first dataset tailored to the preferences of young audiences, enabling the generation of aesthetically oriented dances. And it surpasses the AIST++ dataset in music genre diversity and the intricacy and depth of dance movements. Moreover, the proposed POPDG model within the iDDPM framework enhances dance diversity and, through the Space Augmentation Algorithm, strengthens spatial physical connections between human body joints, ensuring that increased diversity does not compromise generation quality. A streamlined Alignment Module is also designed to improve the temporal alignment between dance and music. Extensive experiments show that POPDG achieves SOTA results on two datasets. Furthermore, the paper also expands on current evaluation metrics. The dataset and code are available at https://github.com/Luke-Luo1/POPDG.

Recently, diffusion-based methods for monocular 3D human pose estimation have achieved state-of-the-art (SOTA) performance by directly regressing the 3D joint coordinates from the 2D pose sequence. Although some methods decompose the task into bone length and bone direction prediction based on the human anatomical skeleton to explicitly incorporate more human body prior constraints, the performance of these methods is significantly lower than that of the SOTA diffusion-based methods. This can be attributed to the tree structure of the human skeleton. Direct application of the disentangled method could amplify the accumulation of hierarchical errors, propagating through each hierarchy. Meanwhile, the hierarchical information has not been fully explored by the previous methods. To address these problems, a Disentangled Diffusion-based 3D Human Pose Estimation method with Hierarchical Spatial and Temporal Denoiser is proposed, termed DDHPose. In our approach: (1) We disentangle the 3D pose and diffuse the bone length and bone direction during the forward process of the diffusion model to effectively model the human pose prior. A disentanglement loss is proposed to supervise diffusion model learning. (2) For the reverse process, we propose Hierarchical Spatial and Temporal Denoiser (HSTDenoiser) to improve the hierarchical modeling of each joint. Our HSTDenoiser comprises two components: the Hierarchical-Related Spatial Transformer (HRST) and the Hierarchical-Related Temporal Transformer (HRTT). HRST exploits joint spatial information and the influence of the parent joint on each joint for spatial modeling, while HRTT utilizes information from both the joint and its hierarchical adjacent joints to explore the hierarchical temporal correlations among joints., Comment: Accepted by AAAI24

Recent work found high mutual information between the learned representations of large language models (LLMs) and the geospatial property of its input, hinting an emergent internal model of space. However, whether this internal space model has any causal effects on the LLMs' behaviors was not answered by that work, led to criticism of these findings as mere statistical correlation. Our study focused on uncovering the causality of the spatial representations in LLMs. In particular, we discovered the potential spatial representations in DeBERTa, GPT-Neo using representational similarity analysis and linear and non-linear probing. Our casual intervention experiments showed that the spatial representations influenced the model's performance on next word prediction and a downstream task that relies on geospatial information. Our experiments suggested that the LLMs learn and use an internal model of space in solving geospatial related tasks., Comment: 12 pages, 15 figures

High content imaging is needed to catalogue the variety of cellular phenotypes and multi-cellular ecosystems present in metazoan tissues. We recently developed Iterative Bleaching Extends multi-pleXity (IBEX), an iterative immunolabeling and chemical bleaching method that enables multiplexed imaging (>65 parameters) in diverse tissues, including human organs relevant for international consortia efforts. IBEX is compatible with over 250 commercially available antibodies, 16 unique fluorophores, and can be easily adopted to different imaging platforms using slides and non-proprietary imaging chambers. The overall protocol consists of iterative cycles of antibody labelling, imaging, and chemical bleaching that can be completed at relatively low cost in 2-5 days by biologists with basic laboratory skills. To support widespread adoption, we provide extensive details on tissue processing, curated lists of validated antibodies, and tissue-specific panels for multiplex imaging. Furthermore, instructions are included on how to automate the method using competitively priced instruments and reagents. Finally, we present a software solution for image alignment that can be executed by individuals without programming experience using open source software and freeware. In summary, IBEX is an open and versatile method that can be readily implemented by academic laboratories and scaled to achieve high content mapping of diverse tissues in support of a Human Reference Atlas or other such applications.

Ample evidence suggests that better machine learning models may be steadily obtained by training on increasingly larger datasets on natural language processing (NLP) problems from non-medical domains. Whether the same holds true for medical NLP has by far not been thoroughly investigated. This work shows that this is indeed not always the case. We reveal the somehow counter-intuitive observation that performant medical NLP models may be obtained with small amount of labeled data, quite the opposite to the common belief, most likely due to the domain specificity of the problem. We show quantitatively the effect of training data size on a fixed test set composed of two of the largest public chest x-ray radiology report datasets on the task of abnormality classification. The trained models not only make use of the training data efficiently, but also outperform the current state-of-the-art rule-based systems by a significant margin., Comment: Accepted at the EMNLP 2020 Clinical NLP workshop, 9 pages + 2 for references, 7 figures, 4 tables

Complex vector light modes, classically-entangled in their spatial and polarisation degrees of freedom (DoF), havebecome ubiquitous in a vast diversity of research fields. Crucially, while polarisation is limited to a bi-dimensionalspace, the spatial mode is unbounded, it can be specified by any of the sets of solutions the wave equation can supportin the different coordinate systems. Here we report on a class of vector beams with elliptical symmetry where thespatial DoF is encoded in the Ince-Gaussian modes of the cylindrical elliptical coordinates. We outline their geometricrepresentation on the Higher-Order Poincar\'e Sphere, demonstrate their experimental generation and analyse the qualityof the generated modes via Stokes polarimetry. We anticipate that such vector modes will be of great relevance inapplications, such as, optical manipulations, laser material processing and optical communications amongst others., Comment: 5 Pages, 6 figures

We study the spatiotemporal correlation of terrorist attacks by al-Qaeda, ISIS, and local insurgents, in six geographical areas identified via $k$-means clustering applied to the Global Terrorism Database. All surveyed organizations exhibit near-repeat activity whereby a prior attack increases the likelihood of a subsequent one by the same group within 20km and on average 4 (al Qaeda) to 10 (ISIS) weeks. Near-response activity, whereby an attack by a given organization elicits further attacks from a different one, is found to depend on the adversarial, neutral or collaborative relationship between the two. When in conflict, local insurgents respond quickly to attacks by global terror groups while global terror groups delay their responses to local insurgents, leading to an asymmetric dynamic. When neutral or allied, attacks by one group enhance the response likelihood of the other, regardless of hierarchy. These trends arise consistently in all clusters for which data is available. Government intervention and spill-over effects are also discussed; we find no evidence of outbidding. Understanding the regional dynamics of terrorism may be greatly beneficial in policy-making and intervention design., Comment: 7-page main article with 4 figures and 14-page supplementary information with 11 figures

Bootstrapping labels from radiology reports has become the scalable alternative to provide inexpensive ground truth for medical imaging. Because of the domain specific nature, state-of-the-art report labeling tools are predominantly rule-based. These tools, however, typically yield a binary 0 or 1 prediction that indicates the presence or absence of abnormalities. These hard targets are then used as ground truth to train image models in the downstream, forcing models to express high degree of certainty even on cases where specificity is low. This could negatively impact the statistical efficiency of image models. We address such an issue by training a Bidirectional Long-Short Term Memory Network to augment heuristic-based discrete labels of X-ray reports from all body regions and achieve performance comparable or better than domain-specific NLP, but with additional uncertainty estimates which enable finer downstream image model training., Comment: Med-NeurIPS-2019 (accepted)

Naive human T cells are produced in the thymus, which atrophies abruptly and severely in response to physical or psychological stress. To understand how an instance of stress affects the size and "diversity" of the peripheral naive T cell pool, we derive a mean-field autonomous ODE model of T cell replenishment that allows us to track the clone abundance distribution (the mean number of different TCRs each represented by a specific number of cells). We identify equilibrium solutions that arise at different rates of T cell production, and derive analytic approximations to the dominant eigenvalues and eigenvectors of the problem linearized about these equilibria. From the forms of the eigenvalues and eigenvectors, we estimate rates at which counts of clones of different sizes converge to and depart from equilibrium values--that is, how the number of clones of different sizes "adjust" to the changing rate of T cell production. Under most physiologically realistic realizations of our model, the dominant eigenvalue (representing the slowest dynamics of the clone abundance distribution) scales as a power law in the thymic output for low output levels, but saturates at higher T cell production rates. Our analysis provides a framework for quantitatively understanding how the clone abundance distributions evolve under small changes in the overall T cell production rate by the thymus., Comment: 22 pages, 7 figures

Acquiring high-quality annotations in medical imaging is usually a costly process. Automatic label extraction with natural language processing (NLP) has emerged as a promising workaround to bypass the need of expert annotation. Despite the convenience, the limitation of such an approximation has not been carefully examined and is not well understood. With a challenging set of 1,000 chest X-ray studies and their corresponding radiology reports, we show that there exists a surprisingly large discrepancy between what radiologists visually perceive and what they clinically report. Furthermore, with inherently flawed report as ground truth, the state-of-the-art medical NLP fails to produce high-fidelity labels., Comment: Accepted workshop contribution for Medical Imaging with Deep Learning (MIDL), 2019

This work provides a strong baseline for the problem of multi-source multi-target domain adaptation and generalization in medical imaging. Using a diverse collection of ten chest X-ray datasets, we empirically demonstrate the benefits of training medical imaging deep learning models on varied patient populations for generalization to out-of-sample domains., Comment: Extended abstract of a journal submission

The rapid spread of radical ideologies has led to a world-wide succession of terrorist attacks in recent years. Understanding how extremist tendencies germinate, develop, and drive individuals to action is important from a cultural standpoint, but also to help formulate response and prevention strategies. Demographic studies, interviews with radicalized subjects, analysis of terrorist databases, reveal that the path to radicalization occurs along progressive steps, where age, social context and peer-to-peer exchanges play major roles. To execute terrorist attacks, radicals must efficiently communicate with one another while maintaining secrecy; they are also subject to pressure from counter-terrorism agencies, public opinion and the need for material resources. Similarly, government entities must gauge which intervention methods are most effective. While a complete understanding of the processes that lead to extremism and violence, and of which deterrents are optimal, is still lacking, mathematical modelers have contributed to the discourse by using tools from statistical mechanics and applied mathematics to describe existing and novel paradigms, and to propose novel counter-terrorism strategies. We review some of their approaches in this work, including compartment models for populations of increasingly extreme views, continuous time models for age-structured radical populations, radicalization as social contagion processes on lattices and social networks, agent based models, game theoretic formulations. We highlight the useful insights offered by analyzing radicalization and terrorism through quantitative frameworks. Finally, we discuss the role of institutional intervention and the stages at which de-radicalization strategies might be most effective., Comment: 56 pages, 12 figures

Successfully integrating newcomers into native communities has become a key issue for policy makers, as the growing number of migrants has brought cultural diversity, new skills, and at times, societal tensions to receiving countries. We develop an agent-based network model to study interacting "hosts" and "guests" and identify the conditions under which cooperative/integrated or uncooperative/segregated societies arise. Players are assumed to seek socioeconomic prosperity through game theoretic rules that shift network links, and cultural acceptance through opinion dynamics. We find that the main predictor of integration under given initial conditions is the timescale associated with cultural adjustment relative to social link remodeling, for both guests and hosts. Fast cultural adjustment results in cooperation and the establishment of host-guest connections that are sustained over long times. Conversely, fast social link remodeling leads to the irreversible formation of isolated enclaves, as migrants and natives optimize their socioeconomic gains through in-group connections. We discuss how migrant population sizes and increasing socioeconomic rewards for host-guest interactions, through governmental incentives or by admitting migrants with highly desirable skills, may affect the overall immigrant experience., Comment: 18 pages, 9 figures

The human adaptive immune response is known to weaken in advanced age, resulting in increased severity of pathogen-born illness, poor vaccine efficacy, and a higher prevalence of cancer in the elderly. Age-related erosion of the T-cell compartment has been implicated as a likely cause, but the underlying mechanisms driving this immunosenescence have not been quantitatively modeled and systematically analyzed. T-cell receptor diversity, or the extent of pathogen-derived antigen responsiveness of the T-cell pool, is known to diminish with age, but inherent experimental difficulties preclude accurate analysis on the full organismal level. In this paper, we formulate a mechanistic mathematical model of T-cell population dynamics on the immunoclonal subpopulation level, which provides quantitative estimates of diversity. We define different estimates for diversity that depend on the individual number of cells in a specific immunoclone. We show that diversity decreases with age primarily due to diminished thymic output of new T-cells and the resulting overall loss of small immunoclones., Comment: 29 pages, 6 figures

Obtaining datasets labeled to facilitate model development is a challenge for most machine learning tasks. The difficulty is heightened for medical imaging, where data itself is limited in accessibility and labeling requires costly time and effort by trained medical specialists. Medical imaging studies, however, are often accompanied by a medical report produced by a radiologist, identifying important features on the corresponding scan for other physicians not specifically trained in radiology. We propose a methodology for approximating image-level labels for radiology studies from associated reports using a general purpose language processing tool for medical concept extraction and sentiment analysis, and simple manually crafted heuristics for false positive reduction. Using this approach, we label more than 175,000 Head CT studies for the presence of 33 features indicative of 11 clinically relevant conditions. For 27 of the 30 keywords that yielded positive results (3 had no occurrences), the lower bound of the confidence intervals created to estimate the percentage of accurately labeled reports was above 85%, with the average being above 95%. Though noisier then manual labeling, these results suggest this method to be a viable means of labeling medical images at scale.

Disaffected youth are among the most susceptible in espousing and acting on extremist ideals, as confirmed by demographic studies. To study age-dependent radicalization we introduce a three-stage model where individuals progress through non-radical, activist, and radical states, while also aging. Transitions between stages are modeled as age-dependent interactions that are maximized for individuals of the same age and that are enhanced at early adulthood. For comparison, we also derive the age-independent formulation corresponding to the full age-dependent model. We find that age-dependence leads to more complex dynamics, enhancing radicalization in certain parameter regimes. We also observe waves of radical behavior ebbing and flowing over generational cycles, realizing well known paradigms in political science. While government intervention is most effective when the appropriate ages are targeted, deciding whether preventive or corrective action is preferable depends on the aggressiveness of the radicalization process.

We introduce a two-variable model to describe spatial polarization, radicalization, and conflict. Individuals in the model harbor a continuous belief variable as well as a discrete radicalization level expressing their tolerance to neighbors with different beliefs. A novel feature of our model is that it incorporates a bistable radicalization process to address memory-dependent social behavior. We demonstrate how bistable radicalization may explain contradicting observations regarding whether social segregation exacerbates or alleviates conflicts. We also extend our model by introducing a mechanism to include institutional influence, such as propaganda or education, and examine its effectiveness. In some parameter regimes, institutional influence may suppress the progression of radicalization and allow a population to achieve social conformity over time. In other cases, institutional intervention may exacerbate the spread of radicalization through a population of mixed beliefs. In such instances, our analysis implies that social segregation may be a viable option against sectarian conflict.

Image deblurring, a.k.a. image deconvolution, recovers a clear image from pixel superposition caused by blur degradation. Few deep convolutional neural networks (CNN) succeed in addressing this task. In this paper, we first demonstrate that the minimum-mean-square-error (MMSE) solution to image deblurring can be interestingly unfolded into a series of residual components. Based on this analysis, we propose a novel iterative residual deconvolution (IRD) algorithm. Further, IRD motivates us to take one step forward to design an explicable and effective CNN architecture for image deconvolution. Specifically, a sequence of residual CNN units are deployed, whose intermediate outputs are then concatenated and integrated, resulting in concatenated residual convolutional network (CRCNet). The experimental results demonstrate that proposed CRCNet not only achieves better quantitative metrics but also recovers more visually plausible texture details compared with state-of-the-art methods., Comment: rejected by AAAI 2019

Diagnostic imaging often requires the simultaneous identification of a multitude of findings of varied size and appearance. Beyond global indication of said findings, the prediction and display of localization information improves trust in and understanding of results when augmenting clinical workflow. Medical training data rarely includes more than global image-level labels as segmentations are time-consuming and expensive to collect. We introduce an approach to managing these practical constraints by applying a novel architecture which learns at multiple resolutions while generating saliency maps with weak supervision. Further, we parameterize the Log-Sum-Exp pooling function with a learnable lower-bounded adaptation (LSE-LBA) to build in a sharpness prior and better handle localizing abnormalities of different sizes using only image-level labels. Applying this approach to interpreting chest x-rays, we set the state of the art on 9 abnormalities in the NIH's CXR14 dataset while generating saliency maps with the highest resolution to date., Comment: submitted to ECCV 2018

The field of medical diagnostics contains a wealth of challenges which closely resemble classical machine learning problems; practical constraints, however, complicate the translation of these endpoints naively into classical architectures. Many tasks in radiology, for example, are largely problems of multi-label classification wherein medical images are interpreted to indicate multiple present or suspected pathologies. Clinical settings drive the necessity for high accuracy simultaneously across a multitude of pathological outcomes and greatly limit the utility of tools which consider only a subset. This issue is exacerbated by a general scarcity of training data and maximizes the need to extract clinically relevant features from available samples -- ideally without the use of pre-trained models which may carry forward undesirable biases from tangentially related tasks. We present and evaluate a partial solution to these constraints in using LSTMs to leverage interdependencies among target labels in predicting 14 pathologic patterns from chest x-rays and establish state of the art results on the largest publicly available chest x-ray dataset from the NIH without pre-training. Furthermore, we propose and discuss alternative evaluation metrics and their relevance in clinical practice., Comment: include the link for the dataset split

Chance constrained optimal power flow (OPF) has been recognized as a promising framework to manage the risk from variable renewable energy (VRE). In presence of VRE uncertainties, this paper discusses a distributionally robust chance constrained approximate AC-OPF. The power flow model employed in the proposed OPF formulation combines an exact AC power flow model at the nominal operation point and an approximate linear power flow model to reflect the system response under uncertainties. The ambiguity set employed in the distributionally robust formulation is the Wasserstein ball centered at the empirical distribution. The proposed OPF model minimizes the expectation of the quadratic cost function w.r.t. the worst-case probability distribution and guarantees the chance constraints satisfied for any distribution in the ambiguity set. The whole method is data-driven in the sense that the ambiguity set is constructed from historical data without any presumption on the type of the probability distribution, and more data leads to smaller ambiguity set and less conservative strategy. Moreover, special problem structures of the proposed problem formulation are exploited to develop an efficient and scalable solution approach. Case studies are carried out on IEEE 14 and 118 bus systems to show the accuracy and necessity of the approximate AC model and the attractive features of the distributionally robust optimization approach compared with other methods to deal with uncertainties.

Most blind deconvolution methods usually pre-define a large kernel size to guarantee the support domain. Blur kernel estimation error is likely to be introduced, yielding severe artifacts in deblurring results. In this paper, we first theoretically and experimentally analyze the mechanism to estimation error in oversized kernel, and show that it holds even on blurry images without noises. Then to suppress this adverse effect, we propose a low rank-based regularization on blur kernel to exploit the structural information in degraded kernels, by which larger-kernel effect can be effectively suppressed. And we propose an efficient optimization algorithm to solve it. Experimental results on benchmark datasets show that the proposed method is comparable with the state-of-the-arts by accordingly setting proper kernel size, and performs much better in handling larger-size kernels quantitatively and qualitatively. The deblurring results on real-world blurry images further validate the effectiveness of the proposed method., Comment: Accepted by WACV 2019

Theano is a Python library that allows to define, optimize, and evaluate mathematical expressions involving multi-dimensional arrays efficiently. Since its introduction, it has been one of the most used CPU and GPU mathematical compilers - especially in the machine learning community - and has shown steady performance improvements. Theano is being actively and continuously developed since 2008, multiple frameworks have been built on top of it and it has been used to produce many state-of-the-art machine learning models. The present article is structured as follows. Section I provides an overview of the Theano software and its community. Section II presents the principal features of Theano and how to use them, and compares them with other similar projects. Section III focuses on recently-introduced functionalities and improvements. Section IV compares the performance of Theano against Torch7 and TensorFlow on several machine learning models. Section V discusses current limitations of Theano and potential ways of improving it., Comment: 19 pages, 5 figures

We derive from first principles a three-dimensional theory of self-propelled particle swarming in a viscous fluid environment. Our model predicts emergent collective behavior that depends critically on fluid opacity, mechanism of self-propulsion, and type of particle-particle interaction. In "clear fluids" swimmers have full knowledge of their surroundings and can adjust their velocities with respect to the lab frame, while in "opaque fluids," they control their velocities only in relation to the local fluid flow. We also show that "social" interactions that affect only a particle's propensity to swim towards or away from neighbors induces a flow field that is qualitatively different from the long-ranged flow fields generated by direct "physical" interactions. The latter can be short-ranged but lead to much longer-ranged fluid-mediated hydrodynamic forces, effectively amplifying the range over which particles interact. These different fluid flows conspire to profoundly affect swarm morphology, kinetically stabilizing or destabilizing swarm configurations that would arise in the absence of fluid. Depending upon the overall interaction potential, the mechanism of swimming (e.g., pushers or pullers), and the degree of fluid opaqueness, we discover a number of new collective three-dimensional patterns including flocks with prolate or oblate shapes, recirculating peloton-like structures, and jet-like fluid flows that entrain particles mediating their escape from the center of mill-like structures. Our results reveal how the interplay among general physical elements influence fluid-mediated interactions and the self-organization, mobility, and stability of new three-dimensional swarms and suggest how they might be used to kinetically control their collective behavior.

We propose an approach to learn spatio-temporal features in videos from intermediate visual representations we call "percepts" using Gated-Recurrent-Unit Recurrent Networks (GRUs).Our method relies on percepts that are extracted from all level of a deep convolutional network trained on the large ImageNet dataset. While high-level percepts contain highly discriminative information, they tend to have a low-spatial resolution. Low-level percepts, on the other hand, preserve a higher spatial resolution from which we can model finer motion patterns. Using low-level percepts can leads to high-dimensionality video representations. To mitigate this effect and control the model number of parameters, we introduce a variant of the GRU model that leverages the convolution operations to enforce sparse connectivity of the model units and share parameters across the input spatial locations. We empirically validate our approach on both Human Action Recognition and Video Captioning tasks. In particular, we achieve results equivalent to state-of-art on the YouTube2Text dataset using a simpler text-decoder model and without extra 3D CNN features., Comment: ICLR 2016

The task of associating images and videos with a natural language description has attracted a great amount of attention recently. Rapid progress has been made in terms of both developing novel algorithms and releasing new datasets. Indeed, the state-of-the-art results on some of the standard datasets have been pushed into the regime where it has become more and more difficult to make significant improvements. Instead of proposing new models, this work investigates the possibility of empirically establishing performance upper bounds on various visual captioning datasets without extra data labelling effort or human evaluation. In particular, it is assumed that visual captioning is decomposed into two steps: from visual inputs to visual concepts, and from visual concepts to natural language descriptions. One would be able to obtain an upper bound when assuming the first step is perfect and only requiring training a conditional language model for the second step. We demonstrate the construction of such bounds on MS-COCO, YouTube2Text and LSMDC (a combination of M-VAD and MPII-MD). Surprisingly, despite of the imperfect process we used for visual concept extraction in the first step and the simplicity of the language model for the second step, we show that current state-of-the-art models fall short when being compared with the learned upper bounds. Furthermore, with such a bound, we quantify several important factors concerning image and video captioning: the number of visual concepts captured by different models, the trade-off between the amount of visual elements captured and their accuracy, and the intrinsic difficulty and blessing of different datasets., Comment: BMVC2016 (Oral paper)

We introduce a novel training principle for probabilistic models that is an alternative to maximum likelihood. The proposed Generative Stochastic Networks (GSN) framework is based on learning the transition operator of a Markov chain whose stationary distribution estimates the data distribution. Because the transition distribution is a conditional distribution generally involving a small move, it has fewer dominant modes, being unimodal in the limit of small moves. Thus, it is easier to learn, more like learning to perform supervised function approximation, with gradients that can be obtained by back-propagation. The theorems provided here generalize recent work on the probabilistic interpretation of denoising auto-encoders and provide an interesting justification for dependency networks and generalized pseudolikelihood (along with defining an appropriate joint distribution and sampling mechanism, even when the conditionals are not consistent). We study how GSNs can be used with missing inputs and can be used to sample subsets of variables given the rest. Successful experiments are conducted, validating these theoretical results, on two image datasets and with a particular architecture that mimics the Deep Boltzmann Machine Gibbs sampler but allows training to proceed with backprop, without the need for layerwise pretraining., Comment: arXiv admin note: substantial text overlap with arXiv:1306.1091

Recent progress in using recurrent neural networks (RNNs) for image description has motivated the exploration of their application for video description. However, while images are static, working with videos requires modeling their dynamic temporal structure and then properly integrating that information into a natural language description. In this context, we propose an approach that successfully takes into account both the local and global temporal structure of videos to produce descriptions. First, our approach incorporates a spatial temporal 3-D convolutional neural network (3-D CNN) representation of the short temporal dynamics. The 3-D CNN representation is trained on video action recognition tasks, so as to produce a representation that is tuned to human motion and behavior. Second we propose a temporal attention mechanism that allows to go beyond local temporal modeling and learns to automatically select the most relevant temporal segments given the text-generating RNN. Our approach exceeds the current state-of-art for both BLEU and METEOR metrics on the Youtube2Text dataset. We also present results on a new, larger and more challenging dataset of paired video and natural language descriptions., Comment: Accepted to ICCV15. This version comes with code release and supplementary material

Neural Autoregressive Distribution Estimators (NADEs) have recently been shown as successful alternatives for modeling high dimensional multimodal distributions. One issue associated with NADEs is that they rely on a particular order of factorization for $P(\mathbf{x})$. This issue has been recently addressed by a variant of NADE called Orderless NADEs and its deeper version, Deep Orderless NADE. Orderless NADEs are trained based on a criterion that stochastically maximizes $P(\mathbf{x})$ with all possible orders of factorizations. Unfortunately, ancestral sampling from deep NADE is very expensive, corresponding to running through a neural net separately predicting each of the visible variables given some others. This work makes a connection between this criterion and the training criterion for Generative Stochastic Networks (GSNs). It shows that training NADEs in this way also trains a GSN, which defines a Markov chain associated with the NADE model. Based on this connection, we show an alternative way to sample from a trained Orderless NADE that allows to trade-off computing time and quality of the samples: a 3 to 10-fold speedup (taking into account the waste due to correlations between consecutive samples of the chain) can be obtained without noticeably reducing the quality of the samples. This is achieved using a novel sampling procedure for GSNs called annealed GSN sampling, similar to tempering methods that combines fast mixing (obtained thanks to steps at high noise levels) with accurate samples (obtained thanks to steps at low noise levels)., Comment: ECML/PKDD 2014

Training of the neural autoregressive density estimator (NADE) can be viewed as doing one step of probabilistic inference on missing values in data. We propose a new model that extends this inference scheme to multiple steps, arguing that it is easier to learn to improve a reconstruction in $k$ steps rather than to learn to reconstruct in a single inference step. The proposed model is an unsupervised building block for deep learning that combines the desirable properties of NADE and multi-predictive training: (1) Its test likelihood can be computed analytically, (2) it is easy to generate independent samples from it, and (3) it uses an inference engine that is a superset of variational inference for Boltzmann machines. The proposed NADE-k is competitive with the state-of-the-art in density estimation on the two datasets tested., Comment: Accepted at Neural Information Processing Systems (NIPS) 2014

Generative Stochastic Networks (GSNs) have been recently introduced as an alternative to traditional probabilistic modeling: instead of parametrizing the data distribution directly, one parametrizes a transition operator for a Markov chain whose stationary distribution is an estimator of the data generating distribution. The result of training is therefore a machine that generates samples through this Markov chain. However, the previously introduced GSN consistency theorems suggest that in order to capture a wide class of distributions, the transition operator in general should be multimodal, something that has not been done before this paper. We introduce for the first time multimodal transition distributions for GSNs, in particular using models in the NADE family (Neural Autoregressive Density Estimator) as output distributions of the transition operator. A NADE model is related to an RBM (and can thus model multimodal distributions) but its likelihood (and likelihood gradient) can be computed easily. The parameters of the NADE are obtained as a learned function of the previous state of the learned Markov chain. Experiments clearly illustrate the advantage of such multimodal transition distributions over unimodal GSNs., Comment: 7 figures, 9 pages, submitted to ICLR14

Several interesting generative learning algorithms involve a complex probability distribution over many random variables, involving intractable normalization constants or latent variable normalization. Some of them may even not have an analytic expression for the unnormalized probability function and no tractable approximation. This makes it difficult to estimate the quality of these models, once they have been trained, or to monitor their quality (e.g. for early stopping) while training. A previously proposed method is based on constructing a non-parametric density estimator of the model's probability function from samples generated by the model. We revisit this idea, propose a more efficient estimator, and prove that it provides a lower bound on the true test log-likelihood, and an unbiased estimator as the number of generated samples goes to infinity, although one that incorporates the effect of poor mixing. We further propose a biased variant of the estimator that can be used reliably with a finite number of samples for the purpose of model comparison., Comment: 10 pages, 1 figure, 2 tables. International Conference on Learning Representations (ICLR'2014, conference track)

Recent work has shown how denoising and contractive autoencoders implicitly capture the structure of the data-generating density, in the case where the corruption noise is Gaussian, the reconstruction error is the squared error, and the data is continuous-valued. This has led to various proposals for sampling from this implicitly learned density function, using Langevin and Metropolis-Hastings MCMC. However, it remained unclear how to connect the training procedure of regularized auto-encoders to the implicit estimation of the underlying data-generating distribution when the data are discrete, or using other forms of corruption process and reconstruction errors. Another issue is the mathematical justification which is only valid in the limit of small corruption noise. We propose here a different attack on the problem, which deals with all these issues: arbitrary (but noisy enough) corruption, arbitrary reconstruction loss (seen as a log-likelihood), handling both discrete and continuous-valued variables, and removing the bias due to non-infinitesimal corruption noise (or non-infinitesimal contractive penalty).

We study a class of swarming problems wherein particles evolve dynamically via pairwise interaction potentials and a velocity selection mechanism. We find that the swarming system undergoes various changes of state as a function of the self-propulsion and interaction potential parameters. In this paper, we utilize a procedure which, in a definitive way, connects a class of individual-based models to their continuum formulations and determine criteria for the validity of the latter. H-stability of the interaction potential plays a fundamental role in determining both the validity of the continuum approximation and the nature of the aggregation state transitions. We perform a linear stability analysis of the continuum model and compare the results to the simulations of the individual-based one.

Time-dependent wave packet calculations for the reaction H+HCl and its isotopic reactions are carried out on the potential energy surface (PES) of Bian and Werner (BW2) [Bian, W.; Werner, H. -J., J. Chem. Phys. 2000, 112, 220]. Reaction probabilities for the exchanged and abstraction channels are calculated from various initial rotational states of the reagent. Those have then been used to estimate reaction cross sections and rate constants which also are calculated and explained by the zero-point energy and the tunneling effect. The results of this work were compared with that of previous quasiclassical trajectory calculations and reaction dynamics experiments on the abstraction channel. In addition, the calculated rate constants are in reasonably good agreement with experimental measurements for both channels. Key words: ab initio, H/D+ DCl/HCl, collision energy, integral cross section, rate constan

The paper presents a theoretical study of the dynamics of the H + HCl system on the potential energy surface (PES) of Bian and Werner (Bian, W.; Werner, H. -J., J. Chem. Phys. 2000, 112, 220). A time-dependent wave packet approach was employed to calculate state-to-state reaction probabilities for the exchanged and abstraction channels. The most recent PES for the system has been used in the calculations. Reaction probabilities have also been calculated for several values of the total angular momentum J > 0. Those have then been used to estimate cross sections and rate constants for both channels. The calculated cross sections can be compared with the results of previous quasiclassical trajectory calculations and reaction dynamics experimental on the abstraction channel. In addition, the calculated rate constants are in the reasonably good agreement with experimental measurement.

Global reservoir information can not only benefit local water management but can also improve our understanding of the hydrological cycle. This information includes water area, elevation, and storage; evaporation rate and volume values; and other characteristics. However, operational wall-to-wall reservoir storage and evaporation monitoring information is lacking on a global scale. Here we introduce NASA’s new MODIS/VIIRS Global Water Reservoir product suite based on moderate resolution remote sensing data—the Moderate Resolution Imaging Spectroradiometer (MODIS), and the Visible Infrared Imaging Radiometer Suite (VIIRS). This product consists of 8-day (MxD28C2 and VNP28C2) and monthly (MxD28C3 and VNP28C3) measurements for 164 large reservoirs (MxD stands for the product from both Terra (MOD) or Aqua (MYD) satellites). The 8-day product provides area, elevation, and storage values, which were generated by first extracting water areas from surface reflectance data and then applying the area estimations to the pre-established Area–Elevation (A–E) relationships. These values were then further aggregated to monthly, with the evaporation rate and volume information added. The evaporation rate and volume values were calculated after the Lake Temperature and Evaporation Model (LTEM) using MODIS/VIIRS land surface temperature product and meteorological data from the Global Land Data Assimilation System (GLDAS). Validation results show that the 250 m area classifications from MODIS agree well with the high-resolution classifications from Landsat (R^(2) = 0.99). Validation of elevation and storage products for twelve Indian reservoirs show good agreement in terms of R^(2) values (0.71–0.96 for elevation, and 0.79–0.96 for storage) and normalized root-mean-square error (NRMSE) values (5.08–19.34% for elevation, and 6.39–18.77% for storage). The evaporation rate results for two reservoirs (Lake Nasser and Lake Mead) agree well with in situ measurements (R^(2) values of 0.61 and 0.66, and NRMSE values of 16.25% and 21.76%). Furthermore, preliminary results from the VIIRS reservoir product have shown good consistency with the MODIS based product, confirming the continuity of this 20-year product suite. This new global water reservoir product suite can provide valuable information with regard to water-sources-related studies, applications, management, and hydrological modeling and change analysis such as drought monitoring.

Knowledge of reservoir bathymetry is essential for many studies on terrestrial hydrological and biogeochemical processes. However, there are currently no cost-effective approaches to derive reservoir bathymetry at the global scale. This study explores the potential of generating high-resolution global bathymetry using elevation data collected by the 532-nm Advanced Topographic Laser Altimeter System (ATLAS) onboard the Ice, Cloud, and Land Elevation Satellite (ICESat-2). The novel algorithm was developed and tested using the ICESat-2 airborne prototype, the Multiple Altimeter Beam Experimental Lidar (MABEL), with Landsat-based water classifications (from 1982 to 2017). MABEL photon elevations were paired with Landsat water occurrence percentiles to establish the elevation-area (E-A) relationship, which in turn was applied to the percentile image to obtain partial bathymetry over the historic dynamic range of reservoir area. The bathymetry for the central area was projected to achieve the full bathymetry. The bathymetry image was then embedded onto the digital elevation model (DEM). Results were validated over Lake Mead against survey data. Results over four transects show coefficient of determination (R²) values from 0.82 to 0.99 and root-mean-square error (RMSE) values from 1.18 to 2.36 m. In addition, the E-A and elevation-storage (E-S) curves have RMSEs of 1.56 m and 0.08 km³, respectively. Over the entire dynamic reservoir area, the derived bathymetry agrees very well with independent survey data, except for within the highest and lowest percentile bands. With abundant overpassing tracks and high spatial resolution, the newly launched ICESat-2 should enable the derivation of bathymetry over an unprecedented number of reservoirs.

The development of compact, high performance Deformable Mirrors (DMs) is one of the most important techno-logical challenges for high-contrast imaging on space missions. Microscale Inc. has fabricated and characterized piezoelectric stack actuator deformable mirrors (PZT-DMs) and Application-Speci c Integrated Circuit (ASIC) drivers for direct integration. The DM-ASIC system is designed to eliminate almost all cables, enabling a very compact optical system with low mass and low power consumption. We report on the optical tests used to evaluate the performance of the DM and ASIC units. We also compare the results to the requirements for space-based high-contrast imaging of exoplanets. .

UNKNOWN