A5077610457- Domain Adaptation and Few-Shot Learning
- Multimodal Machine Learning Applications
- Human Pose and Action Recognition
- Advanced Neural Network Applications
- Advanced Image and Video Retrieval Techniques
- Video Surveillance and Tracking Methods
- 3D Shape Modeling and Analysis
- Machine Learning and ELM
- Adversarial Robustness in Machine Learning
- Advanced Vision and Imaging
- Image Enhancement Techniques
- Computer Graphics and Visualization Techniques
- Advanced Computational Techniques and Applications
- Machine Learning and Algorithms
- Video Analysis and Summarization
- Machine Learning and Data Classification
- Advanced Image Fusion Techniques
- Autonomous Vehicle Technology and Safety
- Reinforcement Learning in Robotics
- Topic Modeling
- Advanced Image Processing Techniques
- Image Retrieval and Classification Techniques
- Face and Expression Recognition
- Analytical Chemistry and Sensors
- Visual Attention and Saliency Detection
Sun Yat-sen University
2013-2024
Zhejiang University
2010-2024
West Virginia University
2024
Shanghai Electric (China)
2024
Beihang University
2018-2024
Vector Institute
2023-2024
Western University
2023-2024
Université Laval
2019-2023
McGill University
2023
Mila - Quebec Artificial Intelligence Institute
2023
In this paper we propose a hybrid architecture of actor-critic algorithms for reinforcement learning in parameterized action space, which consists multiple parallel sub-actor networks to decompose the structured space into simpler spaces along with critic network guide training all networks. While is mainly focused on proposed architecture, call actor-critic, can be extended more general has hierarchical structure. We present an instance based proximal policy optimization (PPO), refer as...
Purpose Although the negative consequences of conflict in work settings have long been recognized, it is only recent years that researchers examined its positive effects, and majority this research has conducted at group level. This paper aims to examine effects on individual behaviors by differentiating between task relationship conflicts, as well moderating influence two contextual variables. Design/methodology/approach A survey was with 166 pairs supervisors subordinates China. Findings...
In this paper, we are proposing a unified and principled method for both the querying training processes in deep batch active learning. We providing theoretical insights from intuition of modeling interactive procedure learning as distribution matching, by adopting Wasserstein distance. As consequence, derived new loss analysis, which is decomposed into optimizing neural network parameters query selection through alternative optimization. addition, naturally formulated min-max optimization...
Multitask learning (MTL) aims at solving the related tasks simultaneously by exploiting shared knowledge to improve performance on individual tasks. Though numerous empirical results supported notion that such among plays an essential role in MTL, theoretical understanding of relationships between and their impact is still open problem. In this work, we are developing a perspective benefits involved using information similarity for MTL. To end, first propose upper bound generalization error...
Remanufacturing is a practice of growing importance due to its increasing environmental and economic benefits. Process planning plays critical role in realizing successful remanufacturing strategy. This paper presents case-based reasoning method for process planning, which allows planner rapidly retrieve, reuse, revise, retain the solutions past problems. In proposed method, influence factors including essential characteristics, failure processing characteristics are identified, local...
Tracking for planar objects is an important issue to vision-based robotic applications. In direct visual tracking (DVT) methods, the similarity between two images often measured through sum of squared differences (SSD) especially with efficient second-order minimization (ESM) due its simplicity and efficiency. However, SSD-based ESM not robust illumination changes since it usually built upon brightness constancy assumption. Contrast image brightness, gradient orientations (GO) are invariant...
Knowing accurate indoor locations of pedestrians has great social and commercial values, such as pedestrian heatmapping targeted advertising. Location estimation with sequential inputs (e.g., geomagnetic sequences) received much attention lately, mainly because they enhance the localization accuracy temporal correlations. Nevertheless, it is challenging to realize sequences due environmental factors, non-uniform ferromagnetic disturbances. To address this, we propose MAIL, a multi-scale...
Comments are used as standard practice in software development to increase the readability of code and express programmers' intentions a more explicit manner. Nevertheless, keeping comments up-to-date is often neglected for programmers. In this paper, we proposed machine learning based method detecting that should be changed during changes. We utilized 64 features, taking before after changes, relationship between into account. Experimental results show 74.6% outdated can detected using our...
Nanoporous noble metals, have shown great advantages due to the high specific surface area and many advanced properties. However, cost is an impediment applications. Looking for low nanoporous metals studying their mechanical properties play important roles. In this paper, inexpensive monolithic copper (NPC) materials are fabricated by dealloying. Microscopically, both of bicontinuous hierarchical structures homogeneous porous at micro/nano scales achieved. The influences dealloying...
Sequences and time-series often arise in robot tasks, e.g., activity recognition imitation learning. In recent years, deep neural networks (DNNs) have emerged as an effective data-driven methodology for processing sequences given sufficient training data compute resources. However, when is limited, simpler models such logic/rule-based methods work surprisingly well, especially relevant prior knowledge applied their construction. unlike DNNs, these "structured" can be difficult to extend, do...
Direct visual tracking (DVT) for planar objects is a fundamental problem in computer vision. DVT methods often formulate as an image registration problem, where intensities are directly used to match two images. However, these usually sensitive illumination changes they assume constant. The gradient orientation (GO) was proven be insensitive variations previous reports. We further confirmed that the GO's robustness can significantly improved when pyramid technique employed. present robust...
Spatiotemporal attention learning remains a challenging video question answering (VideoQA) task as it requires sufficient understanding of cross-modal spatiotemporal information. Existing methods usually leverage different mechanisms to reveal potential associations between and question. While these effectively remove irrelevant information from the attention, they ignore pseudo-related within interaction attention. To address this problem, we proposed novel energy-based refined-attention...