Hengde Zhu

I am a Ph.D. student at School of Computing and Mathematical Sciences, University of Leicester (UoL), United Kingdom, under the supervision of Asst. Prof. Siyang Song, Asst. Prof. Daniel Hao and Prof. Rajeev Raman. I am funded by the Graduate Teaching Assistantship (GTA), University of Leicester. My research interests center on Human-Centrel AI and AI for healthcare. Prior to that, I earned my Master's degree in Human Technology Interaction from UoL, and my Bachelor's degree in Electronic Information Science & Technology from South China Normal University (SCNU).

News

Dec, 2024	One paper on multiple appropriate facial reaction generation has been accepted by AAAI 2025.
July, 2024	One paper on multiple appropriate facial reaction generation has been accepted by ACM MM 2024.

Selected Publications

Google Scholar for all publications

PerReactor: Offline Personalised Multiple Appropriate Facial Reaction Generation
Hengde Zhu, Xiangyu Kong, Weicheng Xie, Xin Huang, Xilin He, Lu Liu, Linlin Shen, Wei Zhang, Hatice Gunes, Siyang Song
AAAI, 2025
[PDF] | [Code]

This paper proposes the first adversarial multiple appropriate facial reaction generation (MAFRG) training strategy which jointly learns appropraiteness and realism discriminators to provide comprehensive task-specific supervision for training the target facial reaction generators, and reformulates the 'one-to-many mapping' training problem as a 'one-to-one (distribution) mapping' training task.

PerFRDiff: Personalised Weight Editing for Multiple Appropriate Facial Reaction Generation
Hengde Zhu*, Xiangyu Kong*, Weicheng Xie, Xin Huang, Linlin Shen, Lu Liu, Hatice Gunes, Siyang Song
ACM Multimedia, 2024
[PDF] | [Code] | [Supplementary Material]

We propose the first online approach for personalized multiple appropriate facial reaction generation (MAFRG) that learns a unique cognitive style from a listener's past facial behaviors and represents it as network weight shifts. These shifts modify a pre-trained MAFRG model, enabling it to mimic the listener's cognitive process in generating appropriate facial reactions (AFRs).

MEEDNets: Medical image classification via ensemble bio-inspired evolutionary DenseNets
Hengde Zhu, Wei Wang, Irek Ulidowski, Qinghua Zhou, Shuihua Wang, Huafeng Chen, Yudong Zhang
Knowledge-Based Systems, 2023
[PDF] | [Code]

This paper proposes an evolutionary mechanism inspired by biological evolution to enhance DenseNet's sparsity and efficiency for medical image classification. A synaptic model mimics asexual reproduction, passing knowledge to descendants while environmental constraints promote sparsity. Additionally, an evolution-based ensemble learning mechanism addresses the need for multiple base networks in decision-making.

An evolutionary attention-based network for medical image classification
Hengde Zhu, Jian Wang, Shuihua Wang, Rajeev Raman, Juan M Górriz, Yudong Zhang
International Journal of Neural Systems, 2023
[PDF] | [Code]

This paper introduces EDCA-Net, an evolutionary attention-based network for robust medical image classification. It builds on DCA-Net, which uses channel-wise weighted feature maps and dense connectivity for efficient information flow. Two evolutionary strategies, intra- and inter-evolution, optimize the network and enhance its generalizability by allowing weight optimization and exchange of training experiences.