A5068530473- Building Energy and Comfort Optimization
- Human Pose and Action Recognition
- Domain Adaptation and Few-Shot Learning
- Advanced Neural Network Applications
- Urban Heat Island Mitigation
- Gait Recognition and Analysis
- Aluminum Alloy Microstructure Properties
- Video Surveillance and Tracking Methods
- Solidification and crystal growth phenomena
- Advanced Image and Video Retrieval Techniques
- Metallurgy and Material Forming
- Solar Radiation and Photovoltaics
- Metallurgical Processes and Thermodynamics
- Impact of Light on Environment and Health
- Anomaly Detection Techniques and Applications
- Manufacturing Process and Optimization
- Multimodal Machine Learning Applications
- Granular flow and fluidized beds
- Machine Learning and Data Classification
- Fluid Dynamics and Heat Transfer
- nanoparticles nucleation surface interactions
- Advanced Numerical Analysis Techniques
- Additive Manufacturing Materials and Processes
- Particle Dynamics in Fluid Flows
- Additive Manufacturing and 3D Printing Technologies
Huazhong University of Science and Technology
2016-2025
Tongji University
2024
Sichuan Agricultural University
2024
Sichuan University
2024
Alibaba Group (China)
2018-2023
State Key Laboratory of Materials Processing and Die & Mould Technology
2019-2023
Alibaba Group (Cayman Islands)
2019-2021
Alibaba Group (United States)
2019-2021
University of Science and Technology of China
2006-2019
South China University of Technology
2019
Although deep neural networks are highly effective, their high computational and memory costs severely hinder applications to portable devices. As a consequence, lowbit quantization, which converts full-precision network into low-bitwidth integer version, has been an active promising research topic. Existing methods formulate the low-bit quantization of as approximation or optimization problem. Approximation-based confront gradient mismatch problem, while optimizationbased only suitable for...
Video-based person re-identification plays an important role in surveillance video analysis, expanding image-based methods by learning features of multiple frames. Most existing fuse temporal average-pooling, without exploring the different frame weights caused various viewpoints, poses, and occlusions. In this paper, we propose attribute-driven method for feature disentangling re-weighting. The single frames are disentangled into groups sub-features, each corresponds to specific semantic...
Cloth-Changing person re-identification (CC-ReID) aims at matching the same across different locations over a long-duration, e.g., days, and therefore inevitably has cases of changing clothing. In this paper, we focus on handling well CC-ReID problem under more challenging setting, i.e., just from single image, which enables an efficient latency-free identity for surveillance. Specifically, introduce Gait recognition as auxiliary task to drive Image ReID model learn cloth-agnostic...
Chemical acute oral toxicity is an important end point in drug design and environmental risk assessment. However, it difficult to determine by experiments, silico methods are hence developed as alternative. In this study, a comprehensive data set containing 12, 204 diverse compounds with median lethal dose (LD₅₀) was compiled. These chemicals were classified into four categories, namely categories I, II, III IV, based on the criterion of U.S. Environmental Protection Agency (EPA). Then...
The goal of gait recognition is to learn the unique spatiotemporal pattern about human body shape from its temporal changing characteristics. As different parts behave differently during walking, it intuitive model spatio-temporal patterns each part separately. However, existing part-based methods equally divide feature maps frame into fixed horizontal stripes get local parts. It obvious that these stripe partition-based cannot accurately locate First, can appear at same (e.g., arms and...
Reactive oxygen species (ROS) act as a group of signaling molecules in rice functioning regulation development and stress responses. Respiratory burst oxidase homologues (Rbohs) are key enzymes generation ROS. However, the role nine Rboh family members was not fully understood multiple disease resistance yield traits. In this study, we constructed mutants each genes detected their requirement Our results revealed that mutations five (RbohA, RbohB, RbohE, RbohH, RbohI) lead to compromised...
Dropout has been proven to be an effective algorithm for training robust deep networks because of its ability prevent overfitting by avoiding the co-adaptation feature detectors. Current explanations dropout include bagging, naive Bayes, regularization, and sex in evolution. According activation patterns neurons human brain, when faced with different situations, firing rates are random continuous, not binary as current does. Inspired this phenomenon, we extend traditional continuous dropout....
Skeleton-based human action recognition has attracted much attention with the prevalence of accessible depth sensors. Recently, graph convolutional networks (GCNs) have been widely used for this task due to their powerful capability model data. The topology adjacency is a key factor modeling correlations input skeletons. Thus, previous methods mainly focus on design/learning topology. But once learned, only single-scale feature and one transformation exist in each layer networks. Many...
Video-based person re-identification (re-ID) aims at matching the same across video clips. Efficiently exploiting multi-scale fine-grained features while building structural interaction among them is pivotal for its success. In this paper, we propose a hybrid framework, Dense Interaction Learning (DenseIL), that takes principal advantages of both CNN-based and Attention-based architectures to tackle video-based re-ID difficulties. DenseIL contains CNN encoder (DI) decoder. The responsible...
Recent works have shown that person re-identification can be substantially improved by introducing attention mechanisms, which allow learning both global and local representations. However, all these learn features in separate branches. As a consequence, the interaction/boosting of information are not allowed, except final feature embedding layer. In this paper, we propose operations as generic family building blocks for synthesizing any This block inserted into convolutional networks with...
Driven by the success of deep learning, last decade has seen rapid advances in person re-identification (re-ID). Nonetheless, most approaches assume that input is given with fulfillment expectations, while imperfect remains rarely explored to date, which a non-trivial problem since directly apply existing methods without adjustment can cause significant performance degradation. In this paper, we focus on recognizing partial (flawed) assistance proposed Part-Part Correspondence Learning...
In single domain generalization, models trained with data from only one are required to perform well on many unseen domains. this paper, we propose a new model, termed meta convolutional neural network, solve the generalization problem in image recognition. The key idea is decompose features of images into features. Acting as "visual words", defined universal and basic visual elements for representations (like words documents language). Taking reference, compositional operations eliminate...
Abstract Carcinogenicity is one of the most concerned properties chemicals to human health, thus it important identify chemical carcinogenicity as early possible. In this study, 829 diverse compounds with rat were collected from Carcinogenic Potency Database (CPDB). Using six types fingerprints represent molecules, 30 binary and ternary classification models generated predict by five machine learning methods. The evaluated an external validation set containing 87 ISSCAN database. best model...
A city is an aggregate of a huge amount heterogeneous data. However, extracting meaningful values from that data remains challenge. City Brain end-to-end system whose goal to glean irreplaceable big data, specifically videos, with the assistance rapidly evolving artificial intelligence technologies and fast-growing computing capacity. From cognition optimisation, decision-making, search prediction ultimately, intervention, improves way manage city, as well live in it. In this study, authors...