A5077534756- Gait Recognition and Analysis
- Domain Adaptation and Few-Shot Learning
- Multimodal Machine Learning Applications
- Privacy-Preserving Technologies in Data
- COVID-19 diagnosis using AI
- Advanced Image and Video Retrieval Techniques
- Image Retrieval and Classification Techniques
- Video Surveillance and Tracking Methods
- Solar Thermal and Photovoltaic Systems
- Hybrid Renewable Energy Systems
- Human Pose and Action Recognition
- Asian Culture and Media Studies
- IoT and GPS-based Vehicle Safety Systems
- Thermodynamic and Exergetic Analyses of Power and Cooling Systems
- Advanced Neural Network Applications
- Phase Change Materials Research
- Microgrid Control and Optimization
- Integrated Energy Systems Optimization
- Smart Grid Energy Management
Queen Mary University of London
2023-2024
Beijing Normal University - Hong Kong Baptist University United International College
2022
Beijing Normal University
2022
Hong Kong Baptist University
2022
Huazhong University of Science and Technology
2019-2021
Conventional centralized deep learning paradigms are not feasible when data from different sources cannot be shared due to privacy or transmission limitation. To resolve this problem, federated has been introduced transfer knowledge across multiple (clients) with non-shared while optimizing a globally generalized central model (server). Existing mostly focus on transmitting image encoders that take instance-sensitive images as input, making them less generalizable and vulnerable inference...
Conventional centralised deep learning paradigms are not feasible when data from different sources cannot be shared due to privacy or transmission limitation. To resolve this problem, federated has been introduced transfer knowledge across multiple (clients) with non-shared while optimising a globally generalized central model (server). Existing mostly focus on transmitting image encoders that take instance-sensitive images as input, making them less generalizable and vulnerable inference...
Existing skeleton-based action recognition methods typically follow a centralized learning paradigm, which can pose privacy concerns when exposing human-related videos. Federated Learning (FL) has attracted much attention due to its outstanding advantages in privacy-preserving. However, directly applying FL approaches skeleton videos suffers from unstable training. In this paper, we investigate and discover that the heterogeneous human topology graph structure is crucial factor hindering...
Composed image retrieval attempts to retrieve an of interest from gallery images through a composed query reference and its corresponding modified text. It has recently attracted attention due the collaboration information-rich concise language precisely express requirements target images. Most current methods follow supervised learning approach training on costly triplet dataset image, text, image. To avoid difficult to-obtain labeled data, zero-shot (ZS-CIR) been introduced, which aims by...
Existing person re-identification (Re-ID) methods mostly follow a centralised learning paradigm which shares all training data to collection for model learning. This is limited when from different sources cannot be shared due privacy concerns. To resolve this problem, two recent works have introduced decentralised (federated) Re-ID constructing globally generalised (server)without any direct access local nor across source domains (clients). However, these are poor on how adapt the maximise...
Existing skeleton-based action recognition methods typically follow a centralized learning paradigm, which can pose privacy concerns when exposing human-related videos. Federated Learning (FL) has attracted much attention due to its outstanding advantages in privacy-preserving. However, directly applying FL approaches skeleton videos suffers from unstable training. In this paper, we investigate and discover that the heterogeneous human topology graph structure is crucial factor hindering...
Text-image composed retrieval aims to retrieve the target image through query, which is specified in form of an plus some text that describes desired modifications input image. It has recently attracted attention due its ability leverage both information-rich images and concise language precisely express requirements for images. However, robustness these approaches against real-world corruptions or further understanding never been studied. In this paper, we perform first study establish...
Conventional centralised deep learning paradigms are not feasible when data from different sources cannot be shared due to privacy or transmission limitation. To resolve this problem, federated has been introduced transfer knowledge across multiple (clients) with non-shared while optimising a globally generalised central model (server). Existing mostly focus on transferring holistic high-level (such as class) models, which closely related specific objects of interest so may suffer inverse...