Traffic forecasting has emerged as a core component of intelligent transportation systems. However, timely accurate traffic forecasting, especially long-term still remains an open challenge due to the highly nonlinear and dynamic spatial-temporal dependencies flows. In this paper, we propose novel paradigm Spatial-Temporal Transformer Networks (STTNs) that leverages dynamical directed spatial long-range temporal improve accuracy forecasting. Specifically, present new variant graph neural...

Abstract Motivation: Genome-wide association studies (GWAS) have been widely used in discovering the between genotypes and phenotypes. Human genome data contain valuable but highly sensitive information. Unprotected disclosure of such information might put individual’s privacy at risk. It is important to protect human data. Exact logistic regression a bias-reduction method based on penalized likelihood discover rare variants that are associated with disease susceptibility. We propose HEALER...

Tile-based rate adaption can improve the quality of experience (QoE) for adaptive 360-degree video streaming under constrained network conditions, which, however, is a challenging problem due to requirements accurate prediction users' viewports and optimal bitrate allocation tiles. In this paper, we propose strategy that deploys reinforcement learning-based Rate Adaptation with Prediction Tiling streaming, named RAPT360, address these challenges. Specifically, accuracy state-of-the-art...

Appearance-based Gaze Estimation leverages deep neural networks to regress the gaze direction from monocular images and achieve impressive performance. However, its success depends on expensive cumbersome annotation capture. When lacking precise annotation, large domain gap hinders performance of trained models new domains. In this paper, we propose a novel adaptation approach, namely Contrastive Regression Adaptation (CRGA), for generalizing estimation target in an unsupervised manner. CRGA...

There has been a recent surge of interest in introducing transformers to 3D human pose estimation (HPE) due their powerful capabilities modeling long-term dependencies. However, existing transformer-based methods treat body joints as equally important inputs and ignore the prior knowledge skeleton topology self-attention mechanism. To tackle this issue, paper, we propose Pose-Oriented Transformer (POT) with uncertainty guided refinement for HPE. Specifically, first develop novel...

Source-free domain adaptation (SFDA) shows the potential to improve generalizability of deep learning-based face anti-spoofing (FAS) while preserving privacy and security sensitive human faces. However, existing SFDA methods are significantly degraded without accessing source data due inability mitigate identity bias in FAS. In this paper, we propose a novel Domain Adaptation framework for FAS (SDA-FAS) that systematically addresses challenges model pre-training, knowledge adaptation, target...

Differentiable architecture search (DARTS) enables effective neural (NAS) using gradient descent, but suffers from high memory and computational costs. In this paper, we propose a novel approach, namely Partially-Connected DARTS (PC-DARTS), to achieve efficient stable by reducing the channel spatial redundancies of super-network. level, partial connection is presented randomly sample small subset channels for operation selection accelerate process suppress over-fitting Side introduced...

Self-supervised learning based on instance discrimination has shown remarkable progress. In particular, contrastive learning, which regards each image as well its augmentations an individual class and tries to distinguish them from all other images, been verified effective for representation learning. However, conventional does not model the relation between semantically similar samples explicitly. this paper, we propose a general module that considers semantic similarity among images. This...

Generalizable face anti-spoofing (FAS) based on domain generalization (DG) has gained growing attention due to its robustness in real-world applications. However, these DG methods rely heavily labeled source data, which are usually costly and hard access. Comparably, unlabeled data far more accessible various scenarios. In this paper, we propose the first Unsupervised Domain Generalization framework for Face Anti-Spoofing, namely UDG-FAS, could exploit large amounts of easily learn...

The increasing availability of genome data motivates massive research studies in personalized treatment and precision medicine. Public cloud services provide a flexible way to mitigate the storage computation burden conducting genome-wide association (GWAS). However, privacy has been widely concerned when sharing sensitive information environment. We presented novel framework (FORESEE: Fully Outsourced secuRe gEnome Study basEd on homomorphic Encryption) fully outsource GWAS (i.e.,...

In biomedical research, data sharing and information exchange are very important for improving quality of care, accelerating discovery, promoting the meaningful secondary use clinical data. A big concern in is protection patient privacy because inappropriate leakage can put at risk. this study, we deployed a grid logistic regression framework based on Secure Multi-party Computation (SMAC-GLORE). Unlike our previous work GLORE, SMAC-GLORE protects not only patient-level data, but also all...

Privacyis an important consideration when sharing clinical data, which often contain sensitive information. Adequate protection to safeguard patient privacy and increase public trust in biomedical research is paramount. This review covers topics policy technology the context of data sharing. We articles related (

With the advent of data science, analysis network or graph has become a very timely research problem. A variety recent works have been proposed to generalize neural networks graphs, either from spectral theory spatial perspective. The majority these works, however, focus on adapting convolution operator representation. At same time, pooling also plays an important role in distilling multiscale and hierarchical representations, but it mostly overlooked so far. In this article, we propose...

Batch normalization (BN) is a fundamental unit in modern deep neural networks. However, BN and its variants focus on statistics but neglect the recovery step that uses linear transformation to improve capacity of fitting complex data distributions. In this paper, we demonstrate can be improved by aggregating neighborhood each neuron rather than just considering single neuron. Specifically, propose simple yet effective method named batch with enhanced (BNET) embed spatial contextual...

In federated learning (FL), the heterogeneity of data and asynchronous participation clients have been observed to induce local client's model discrepancy with high variance, leading a slow unstable convergence globally at server. this article, motivated by usefulness stale client updates, we first propose general framework, named FedVR, address issue. design an aggregate both fresh updates without additional communication overhead, which is computed server as control variate reduce variance...

Image compression for both human and machine vision has become prevailing to accommodate rising demands machine-machine human-machine communications. Scalable image is recently emerging as an efficient alternative simultaneously achieve high accuracy in the base layer obtain high-fidelity reconstruction enhancement layer. However, existing methods scalable coding with heuristic mechanisms, which cannot fully exploit inter-layer correlations evidently sacrifice rate-distortion performance. In...

Existing methods for integerized training speed up deep learning by using low-bitwidth weights, activations, gradients, and optimizer buffers. However, they overlook the issue of full-precision latent which consume excessive memory to accumulate gradient-based updates optimizing weights. In this paper, we propose first weight quantization schema general training, minimizes perturbation process via residual with optimized dual quantizer. We leverage eliminate correlation between suppressing...

Learned image compression (LIC) has demonstrated superior rate-distortion (R-D) performance compared to traditional codecs, but is challenged by training inefficiency that could incur more than two weeks train a state-of-the-art model from scratch. Existing LIC methods overlook the slow convergence caused compacting energy in learning nonlinear transforms. In this paper, we first reveal such compaction consists of components, i.e., feature decorrelation and uneven modulation. On basis,...