Cheng Wan
Hi there! I’m an ECE Ph.D. student at Cornell University, and I’m honored to be advised by Mert Sabuncu and Qingyu Zhao. In Summer 2026, I was a research intern on the Adobe Firefly foundation model team, working on large-scale pre-training and post-training of generative models. Previously, I worked on computer vision at CMU and on AI for healthcare at Georgia Tech and Emory.
Research Interests: Generative modeling (diffusion / flow matching, image & video generation), multimodal reasoning, and foundation-model pre-/post-training (on-policy distillation, RL), with applications to medical imaging.
Experience
Publications
-
R-OPSD: Realizable On-Policy Self-Distillation for Diffusion ModelsUnder review, 2026 -
AdaOPD: Mitigating Language-Style Gap in On-Policy Distillation for Medical VLM Post-TrainingUnder review, 2026 -
IEEE Trans. on Medical Imaging (TMI) -
FlowMap-OPD: Decoupling Rollout and Supervision for Few-Step Flow-Map GeneratorsUnder review, 2026
Academic Service
Reviewer: NeurIPS (2024–2026), ICLR (2025–2026), ICML (2025), AISTATS (2025–2026)
Workshop Reviewer: NeurIPS AI4Science (2025–2026), NeurIPS SPIGM (2025–2026), ICML AI4Science (2024)
Teaching
- Teaching Assistant (Co-lecturer), ECE 5200/3200: Foundations of Machine Learning, Cornell University Spring 2025