Chen Wei

I am an Assistant Professor of Computer Science at Rice University.

I completed my Ph.D. at Johns Hopkins University, advised by Bloomberg Distinguished Professor Alan L. Yuille. After my PhD, I joined Meta FAIR as a postdoctoral researcher. During my doctoral studies, I also interned at Google DeepMind and Meta FAIR. I received my B.S. with honors from Peking University EECS.

Research

My research asks: how can AI learn to see, reason about, and act in the visual world with minimal supervision? I approach this from three angles:

Visual Representation Learning: building foundational visual encoders (iBOT, MaskFeat, Perception Encoder) that power downstream perception at scale.
Generative World Models: using generative models not just to create, but to understand: learning representations from diffusion (DiffMAE, De-Diffusion, DSR) and generating explorable worlds (GenEx).
Agentic Vision: enabling models to reason and act through vision, from game-based reasoning (Play to Generalize) to tool-augmented visual agents (PyVision, PyVision-RL).

News

2026 Serving as Area Chair for NeurIPS 2026.
2026 Congratulations to Shitian Zhao and Yunfei Xie — PyVision-RL and MEMO accepted to ICML 2026!
2026 Co-organizing the CVPR 2026 Workshop on World Models Meet Active Sensing and Closed-Loop Planning.
2026 Selected as an AAAI New Faculty Highlight.
2026 Congratulations to Yunfei Xie — Play to Generalize accepted to ICLR 2026!
2025 Perception Encoder accepted to NeurIPS 2025 as Oral presentation.
2025 Congratulations to Yunfei Xie for receiving the Lambda Research Grant ($10,000) for his work on Play to Generalize.

Blog

with Yifei Wang and Xiaoyu Wu · June 2026 · generative model evaluation

EMA: A Quiet Hyperparameter That Moves Diffusion Leaderboards

We examine exponential moving average (EMA), an often overlooked hyperparameter, in recent speed-run image diffusion benchmarks. A seemingly small experimental detail can change how benchmark results and model rankings should be interpreted.

Selected Publications

Taming Outlier Tokens in Diffusion Transformers
Xiaoyu Wu*, Yifei Wang*, Tsu-Jui Fu, Liang-Chieh Chen, Zhe Gan, Chen Wei
arXiv preprint, 2026
arXiv / project page / code

MEMO: Memory-Augmented Model Context Optimization for Robust Multi-Turn Multi-Agent LLM Games
Yunfei Xie, Kevin Wang, Bobby Cheng, Jianzhu Yao, Zhizhou Sha, Alexander Duffy, Yihan Xi, Hongyuan Mei, Cheston Tan, Chen Wei, Pramod Viswanath, Zhangyang Wang
ICML, 2026
arXiv / project page / code

Play to Generalize: Learning to Reason Through Game Play
Yunfei Xie, Yinsong Ma, Shiyi Lan, Alan Yuille, Junfei Xiao, Chen Wei
ICLR, 2026
arXiv / project page / code

PyVision-RL: Forging Open Agentic Vision Models via RL
Shitian Zhao, Shaoheng Lin, Ming Li, Haoquan Zhang, Wenshuo Peng, Kaipeng Zhang†, Chen Wei†
ICML, 2026
arXiv / project page / code

PyVision: Agentic Vision with Dynamic Tooling
Shitian Zhao, Haoquan Zhang, Shaoheng Lin, Ming Li, Qilong Wu, Kaipeng Zhang†, Chen Wei†
NeurIPS 2025 Workshop on Multi-Turn Interactions in LLMs
arXiv / project page / code

GenEx: Generating an Explorable World
Taiming Lu, Tianmin Shu, Junfei Xiao, Luoxin Ye, Jiahao Wang, Cheng Peng, Chen Wei, Daniel Khashabi, Rama Chellappa, Alan Yuille, Jieneng Chen
ICLR, 2025
arXiv / genex.world / code

De-Diffusion Makes Text a Strong Cross-Modal Interface
Chen Wei, Chenxi Liu, Siyuan Qiao, Zhishuai Zhang, Alan Yuille, Jiahui Yu
CVPR, 2024
arXiv / project page

Diffusion Models as Masked Autoencoders
Chen Wei, Karttikeya Mangalam, Po-Yao Huang, Yanghao Li, Haoqi Fan, Hu Xu, Huiyu Wang, Cihang Xie, Alan Yuille, Christoph Feichtenhofer
ICCV, 2023
arXiv / project page / press

Masked Feature Prediction for Self-Supervised Visual Pre-Training
Chen Wei*, Haoqi Fan, Saining Xie, Chao-Yuan Wu, Alan Yuille, Christoph Feichtenhofer*
CVPR, 2022
arXiv / code / Most Influential CVPR 2023 Papers

iBOT: Image BERT Pre-Training with Online Tokenizer
Jinghao Zhou, Chen Wei, Huiyu Wang, Wei Shen, Cihang Xie, Alan Yuille, Tao Kong
ICLR, 2022
arXiv / code / press / Improved and scaled up to DINOv2 by Meta AI.