|
Kaihui Cheng | 成凯回 |
|
Biography
I am a Ph.D. student at Fudan University, supervised by Prof. Siyu Zhu. My research focuses on generative models.
Research Interest
- Generative Models
- 3D Generation
Projects & Publications
AlphaFolding: 4D Diffusion for Dynamic Protein Structure Prediction with Reference and Motion Guidance
Authors: Kaihui Cheng, Chang Liu, Qi Su, Jun Wang, Liwei Zhang, Yining Tang, Yao Yao, Siyu Zhu†, Yuan Qi†
Venue: AAAI 2025
Description: A 4D diffusion model that predicts dynamic protein structures across multiple time steps, jointly modeling backbone and side chains. Trained on molecular dynamics trajectories, it uses reference-structure and motion guidance to generate temporally consistent conformational dynamics.
Dynamic PDB: A New Dataset and a SE(3) Model Extension by Integrating Dynamic Behaviors and Physical Properties in Protein Structures
Authors: C. Liu, J. Wang, Z. Cai, Y. Wang, H. Kuang, Kaihui Cheng, L. Zhang, Q. Su, Y. Tang, et al.
Venue: arXiv 2024
Description: A large-scale dataset of ~12.6K proteins, each with 1 microsecond all-atom MD simulations and a full suite of physical properties (atomic velocities, forces, potential/kinetic energies, temperature) recorded at 1 ps intervals. It also extends an SE(3) diffusion model with these physical properties for trajectory prediction.
Text-Video Retrieval Re-ranking via Multi-grained Cross Attention and Frozen Image Encoders
Authors: Z. Dai, Kaihui Cheng, F. Shao, Z. Dong, S. Zhu†
Venue: Pattern Recognition 2025
Description: A multi-grained cross-attention re-ranker (CrossTVR) for text-video retrieval. Frame- and video-level token selectors extract salient visual tokens, and a cross-attention module is trained on top of a frozen vision backbone, achieving state-of-the-art retrieval with low training overhead and scalability to larger foundation models.