Postdoctoral Scholar · Princeton University
I am currently a Postdoctoral Scholar in the Department of ECE at Princeton University, co-affiliated with Princeton AI Lab, fortunately working with Prof. Mengdi Wang. Prior to this, I served as a Senior Research Assistant at Princeton University from January to July 2025. I received my Ph.D. degree from Peking University in July 2025, jointly supervised by Prof. Bin Cui and Prof. Luxia Zhang. During my doctoral studies, I was selected for the ByteDance Top Seed Talent Program.
I was also fortunate to collaborate with Yang Song, Shuicheng Yan, Ming-Hsuan Yang, Bernard Ghanem, Jiajun Wu, Stefano Ermon, Jure Leskovec, Yejin Choi, and James Zou.
I currently focus on developing advanced generative models, including their training methodologies, architecture design, alignment, inference efficiency and applications. I am in charge of a research team at Princeton and have led a series of works on LLMs/MLLMs and Diffusion Models, including RPG-DiffusionMaster , MMaDA
, Buffer of Thoughts
, ReasonFlux/PRM/Coder
, LatentMAS
, dLLM-RL
, Consistency Flow Matching
, RLAnything
.
We are looking for collaborators for research in LLM/MLLM Post-Training, Diffusion LLMs, World Modeling, and Agent Training. Positions are available across multiple levels and institutions:
We take collaboration seriously β please make sure the following apply before reaching out:
In return, we offer:
My overarching goal is to build a unified system for Physics-Aware Generative Intelligence β advancing AI from passive pattern recognition toward active reasoning, simulation, and discovery in the physical world.
This vision is organized around two complementary pillars. The first, Generative Model Foundations, pursues the core algorithmic and architectural advances that power the next generation of AI β spanning language model reasoning, large-scale reinforcement learning, intelligent agent systems, and diffusion model innovations. The second, Generative Applications, deploys these foundations to tackle real-world challenges: generating coherent multimodal content across image, 3D, and 4D domains, and accelerating scientific discovery through AI-driven hypothesis generation and evaluation.
The two pillars are united by a shared philosophy: theoretical depth should translate directly into empirical impact, and generalization across modalities and scenarios is the ultimate test of any AI system's intelligence.
Buffer of Thought, MMaDA, ReasonFlux, ReasonFlux-Coder, ReasonFlux-PRM, TraDo, SuperCorrect, LatentMAS, AutoTool, RLAnything
dLLM-RL, MMaDA, ReasonFlux-Coder, ReasonFlux-PRM, HermesFlow, Demystifying Agent RL, RLAnything
Alita, AgentDistill, ScoreFlow, Multi-Actor Collaboration, Preacher, EmoAgent, DemyAgent, LatentMAS, GenEnv, RLAnything
RPG, MMaDA, ContextDiff, dLLM-RL, Consistency Flow Matching, Diffusion-Sharpening, Rectified Diffusion, ConPreDiff, SADM, MMaDA-Parallel
IterComp, VideoTetris, ScoreLiDAR, IPDreamer, EditWorld, Trans4D, WideRange4D, OmniVerifier, MMaDA-Parallel, UltraViCo
"Diffusion Model: Theory, Application, and Code Practice of Generative AI Models"
Published by Electronics Industry Press (η΅εε·₯δΈεΊηη€Ύ), 2023 · Purchase Link · Selected as Annual Outstanding Author
*Co-first author, +Corresponding author. For a complete list, see my Google Scholar profile.
10 selected worldwide · 2025
Peking University Ph.D. · 2025
24 selected worldwide · 2025
8 selected in China · 2024
Electronics Industry Press · 2023
Top 1% at PKU · 2022
Top 1% at PKU · 2022
NeurIPS, ICLR