Yifei Wang
Hi, I'm Yifei Wang
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.
News
- new New blog post: EMA: A Quiet Hyperparameter That Moves Diffusion Leaderboards.
- new Joining Prof. Alan Yuille's lab at JHU as a visiting student for the summer.
- new Released DSR (with Apple).
- Uni-Instruct (with Xiaohongshu Inc.) accepted to NeurIPS 2025.
- Started PhD at Rice University, working with Prof. Chen Wei.
- EM-Diffusion accepted to NeurIPS 2024.
Selected work
Taming Outlier Tokens in Diffusion Transformers
Preprint · 2026
Outlier patch tokens hurt both ViT encoders and diffusion transformers in RAE-DiT pipelines. Our Dual-Stage Registers patch both sides and improve ImageNet-256 FID from 5.89 → 4.58 at 80 epochs.
Uni-Instruct: One-step Diffusion through Unified Divergence Instruction
NeurIPS · 2025
A single f-divergence framework that subsumes 10+ one-step diffusion distillation methods (Diff-Instruct, DMD, SiD, SIM, …) — and a new SoTA one-step FID of 1.02 on ImageNet 64×64, beating the 35-NFE EDM teacher.
About
I'm a second-year PhD student at Rice University, where I am working with Prof. Chen Wei. In May 2026 I visited Prof. Alan Yuille's lab at Johns Hopkins University. Before Rice I received my B.S. from Peking University, advised by Prof. He Sun and Dr. Weijian Luo. My research focuses on generative modeling, with an emphasis on the theory and the practical bottlenecks that govern their training.