Full Professor, Hong Kong University of Science and Technology
4 papers at NeurIPS 2025
A novel alignment framework that integrates generative reward models with multi-modal RLHF.
Safe RLHF-V, the multimodal safety alignment framework.
Facial personalization models fail to control facial attributes accurately. FreeCure proposed in this paper fixes this problem with a training-free framework without harming these models' impressive ability in maintaining identity information.
A human preference dataset for multi-turn interleaved multimodal understanding and generatin tasks
