Accurate long-term trajectory prediction in complex scenes, where multiple agents (e.g., pedestrians or vehicles) interact with each other and the environment while attempting to accomplish diverse often unknown goals, is a challenging stochastic forecasting problem. In this work, we propose MUSEVAE, new probabilistic modeling framework based on cascade of Conditional VAEs, which tackles long-term, uncertain task using coarse-to-fine multi-factor architecture. its Macro stage, model learns...

We propose new methods to speed up convergence of the Alternating Direction Method Multipliers (ADMM), a common optimization tool in context large scale and distributed learning. The proposed method accelerates by automatically deciding constraint penalty needed for parameter consensus each iteration. In addition, we also an extension that adaptively determines maximum number iterations update penalty. show this approach effectively leads adaptive, dynamic network topology underlying...

Recent studies in image memorability showed that the of an is a measurable quantity and closely correlated with semantic attributes. However, intrinsic characteristics are not yet fully understood. It has been reported contrast to popular belief unusualness or aesthetic beauty may be positively memorability. This counter-intuitive characteristic hinders better understanding its applicability. In this paper, we investigate two new spatial features intuitively explainable. We propose Weighted...

We propose new methods to speed up convergence of the Alternating Direction Method Multipliers (ADMM), a common optimization tool in context large scale and distributed learning. The proposed method accelerates by automatically deciding constraint penalty needed for parameter consensus each iteration. In addition, we also an extension that adaptively determines maximum number iterations update penalty. show this approach effectively leads adaptive, dynamic network topology underlying...

FSS (Few-shot segmentation) aims to segment a target class using small number of labeled images (support set). To extract information relevant the class, dominant approach in best performing methods removes background features support mask. We observe that this feature excision through limiting mask introduces an bottleneck several challenging cases, e.g., for targets and/or inaccurate boundaries. end, we present novel method (MSI), which maximizes support-set by exploiting two complementary...

Digital mammography is one of the most promising options to diagnose breast cancer which common in women. However, its effectiveness enfeebled due difficulty distinguishing actual lesions from benign abnormalities, results unnecessary biopsy referrals. To overcome this issue, computer aided diagnosis (CADx) using machine learning techniques have been studied worldwide. Since a classification problem and number features obtainable mammogram image infinite, feature selection method that...

Computational analysis and prediction of digital media interestingness is a challenging task, largely driven by subjective nature interestingness. Several attempts were made to construct reliable measure obtain better understanding based on various psychological study results. However, most current works focus for images. While the video affective has been studied quite some time, there are few that explictly try predict videos. In this work, we extend recent pilot using mid-level...

For transportation hubs, leveraging pedestrian flows for commercial activities presents an effective strategy funding maintenance and infrastructure improvements. However, this introduces new challenges, as consumer behaviors can disrupt flow efficiency. To optimize both retail potential efficiency, careful strategic planning in store layout facility dimensions was done by expert judgement due to the complexity dynamics areas of hubs. This paper attention-based movement model simulate these...

Crowd simulations facilitate the study of how an environment layout impacts movement and behavior its inhabitants. However, are computationally expensive, which make them infeasible when used as part interactive systems (e.g., Computer-Assisted Design software). Machine learning models, such neural networks (NN), can learn observed behaviors from examples, potentially offer a rational prediction crowd's efficiently. To this end, we propose method to predict aggregate characteristics crowd...

Bayesian models provide a framework for probabilistic modelling of complex datasets. Many such are computationally demanding, especially in the presence large In sensor network applications, statistical (Bayesian) parameter estimation usually relies on decentralized algorithms, which both data and computation distributed across nodes network. this paper we propose learning using Bregman Alternating Direction Method Multipliers (BADMM). We demonstrate utility our framework, with Mean Field...

Digital mammography is one of the most promising options to diagnose breast cancer which common in women. However, it has weaknesses excessive unnecessary biopsy referrals that enfeeble its effectiveness due difficulty distinguishing actual lesions from benign abnormalities. To overcome this issue, computer aided diagnosis (CADx) using machine learning techniques have been studied worldwide. Since a classification problem and number features obtainable mammogram image theoretically infinite,...

Tracking the movement of individuals in a crowd is an indispensable component to reconstructing movement, with applications surveillance and data-driven animation. Typically, multiple sensors are distributed over wide area often they have incomplete coverage or input introduces noise due tracking algorithm hardware failure. In this paper, we propose novel refinement method that complements existing solutions reconstruct holistic view microscopic crowd, from noisy tracked data missing even...

Computational image memorability prediction has made significant progress in recent years. It is reported that we can robustly estimate the of images with many different object and scene classes. However, large scale data-based method including deep Convolutional Neural Networks (CNNs) showed a room for improvement when it was applied to smaller benchmark dataset. In this work, investigate missing link causes such performance gap via in-depth qualitative analysis, then provide suggestions...

We propose new methods to speed up convergence of the Alternating Direction Method Multipliers (ADMM), a common optimization tool in context large scale and distributed learning. The proposed method accelerates by automatically deciding constraint penalty needed for parameter consensus each iteration. In addition, we also an extension that adaptively determines maximum number iterations update penalty. show this approach effectively leads adaptive, dynamic network topology underlying...

Computer-aided diagnosis of mass lesions in Digital Database for Screening Mammography (DDSM) is investigated using a recently developed SVM based on recursive feature elimination (SVM-RFE) as the classification technique. To evaluate generalizability, computer-aided cross-institutional mammograms also examined. The results this paper indicate that only subset available set features facilitates increased accuracy, and accuracy generally lower than when same-institutional mammograms.

Resource constraints frequently complicate multi-agent planning problems. Existing algorithms for resource-constrained, problems rely on the assumption that are deterministic. However, resource themselves subject to uncertainty from external influences. Uncertainty about is especially challenging when agents must execute in an environment where communication unreliable, making on-line coordination difficult. In those cases, it a significant challenge find coordinated allocations at plan time...

Bayesian models provide a framework for probabilistic modelling of complex datasets. However, many such are computationally demanding especially in the presence large On other hand, sensor network applications, statistical (Bayesian) parameter estimation usually needs distributed algorithms, which both data and computation across nodes network. In this paper we propose general learning using Bregman Alternating Direction Method Multipliers (B-ADMM). We demonstrate utility our framework, with...