Second-year PhD student · Rice University
I work on generative models, including diffusion, flow matching, and the building blocks that make them more efficient and controllable.
At Rice, I am advised by Prof. Chen Wei. My research connects the theory of generative modeling with the practical challenges of training and inference.
Preprint · 2026
Pass@k exposes a quality–coverage tradeoff in diffusion distillation: distribution-matching methods can improve single-draw quality while losing coverage across image and video generation.
Preprint · 2026
Cross-view class-token alignment strengthens both global and patch representations in diffusion transformers while preserving generation quality.
Preprint · 2026
A controlled Rubik's Cube benchmark tests hidden-state reasoning in video models, separating visual plausibility from state accuracy and measuring the compute cost of scaling.
Preprint · 2026
Prediction-Aligned Context Compaction learns memory tokens through on-policy distillation from a frozen video generator, improving long-video memory without updating the generator.
Preprint · 2026
Dual-Stage Registers address outlier tokens in both the encoder and diffusion transformer, improving ImageNet-256 FID from 5.89 to 4.58 at 80 epochs.
NeurIPS 2025
A unified f-divergence framework connecting more than 10 one-step diffusion distillation methods, with a one-step FID of 1.02 on ImageNet 64 × 64.
* Equal contribution.
Before Rice, I received my B.S. from Peking University, advised by Prof. He Sun and Dr. Weijian Luo. In May 2026, I visited Prof. Alan Yuille's lab at Johns Hopkins University.
