Research
Research statement
My research focuses on generative models, especially diffusion models, flow matching, and efficient image generation. I am interested in how modeling choices such as architectures, training objectives, noise schedules, samplers, and distillation shape sample quality, likelihood, and deployment cost.
Recently, I have been working on representation-space diffusion, one-step and few-step generation, and practical design choices behind high-quality text-to-image systems. My goal is to connect the theory of generative modeling with methods that are reliable and efficient in real applications.
List of publications
(CVPR 2026) Masked Auto-Regressive Variational Acceleration: Fast Inference Makes Practical Reinforcement Learning Yuxuan Gu, Weimin Bai, Yifei Wang, Weijian Luo and He Sun. paper | project page
(Preprint 2026) Unbiased Diffusion Variational Inversion via Principled Posterior Matching Weimin Bai, Yuxuan Gu, Yifei Wang, Weijian Luo and He Sun. paper
(Preprint 2026) Taming Outlier Tokens in Diffusion Transformers Xiaoyu Wu*, Yifei Wang*, Tsu-Jui Fu, Liang-Chieh Chen, Zhe Gan and Chen Wei. paper | project page | code
(NeurIPS 2025) Uni-Instruct: One-step Diffusion Model through Unified Diffusion Divergence Instruction Yifei Wang, Weimin Bai, Colin Zhang, Debing Zhang, Weijian Luo and He Sun. paper | project page | code
(NeurIPS 2024) An Expectation-Maximization Algorithm for Training Clean Diffusion Models from Corrupted Observations Weimin Bai, Yifei Wang, Wenzheng Chen and He Sun. paper
(Arxiv preprint) Integrating Amortized Inference with Diffusion Models for Learning Clean Distribution from Corrupted Images Yifei Wang, Weimin Bai, Weijian Luo, Wenzheng Chen and He Sun. paper