PhD Researcher · UCAS
I study generative models, multimodal understanding, and reinforcement learning, with a focus on making visual generation and reasoning more reliable and effective.
Do what’s right.
What I work on
Compositional prototypes and adaptive fusion for transferring reusable component representations to novel visual classes.
Compositional Prototypical NetworksStyle-transfer regularization and collage priors for high-quality image composition across diverse visual domains.
StyleComTree-based trajectories and multi-group advantage estimation for optimizing text-to-image models with multiple rewards.
Multi-GRPOTeacher-supervised consistency rewards for video reasoning under spatial and temporal degradation.
Video-AlignerI am a direct-entry PhD student in Computer Science at the University of Chinese Academy of Sciences. My work spans few-shot learning, reliable video reasoning, text-to-image generation, cross-domain image composition, and efficient video generation.
Across research internships at Alibaba Cloud and Youku, and algorithm internships at RightBrainAI and Megvii, I have worked on problems ranging from multimodal reinforcement learning to controllable video generation. I enjoy turning open-ended research questions into reproducible experiments and practical systems. I expect to complete my PhD in June 2027.
Research output
Notes and explainers
Community
Page views and geographic distribution since September 2026