Ph.D. Candidate @ Nankai University · Qingyun Program Intern @ Tencent Hunyuan
I build multimodal foundation models that can perceive, reason about, and interact with the physical world. I am advised by Prof. Ming-Ming Cheng and Prof. Qibin Hou, and previously worked at ByteDance's Volcano Engine Multimedia Laboratory.
Annotations as Rollouts Efficient and scalable reinforcement learning across seven video-perception task families, without chain-of-thought decoding.
|
Efficient Physical-World Agents An efficient MoE foundation model for action-centric perception, multi-turn interaction, and long-horizon embodied reasoning.
|
NeurIPS 2025 Reinforcement fine-tuning with effective temporal sampling for precise video temporal grounding.
|
Universal Video MLLMs Attribute-structured and quality-verified data and models for fine-grained audiovisual video understanding.
|
See my academic homepage for the full publication list, projects, and resources.
Open to conversations and collaborations on multimodal, embodied, and physical AI.