I am a third-year Ph.D. student at Stanford CS, co-advised by James Zou and Stefano Ermon. I received my B.S. in Math and CS at Yuanpei College, Peking University, co-advised by Liwei Wang and Di He. During my undergrad, I was fortunate to work with Jacob Steinhardt and Simon S. Du as a visiting researcher. I also collaborated with Ming-Yu Liu as an intern in NVIDIA Cosmos, and Felix Yu in Google Deepmind.

My Ph.D. research aims to advance generative AI capabilities for open problems through post-training and inference-time algorithm design. Specifically, for both text and vision generative models:

• How do we continuously improve them via post-training, even at superhuman levels?
• How do we scale test-time compute effectively and efficiently?

I believe in the synergy of scalable engineering and principled algorithm design. This philosophy is grounded in my early research on deep learning foundations and AI for science. Feel free to reach out if you are interested in my research or simply want to chat.

News

• (Jan. 2026) We release an interesting study on whether AI can discover its own scaling laws better than humans. Check it out!
• (Jan. 2026) Check out InfoTok, our adaptive information-theoretic tokenizer that achieves 40% higher compression without information loss and represents videos via coarse-grained to fine-grained sequences.
• (Dec. 2025) We release DDRL, the RL algorithm powering NVIDIA’s Cosmos-Predict2.5 video generative models (2B/14B)! An amazing experience scaling the post-training of models on ~2,000 H100 GPUs. Check our post and models!
• (Dec. 2025) Data Attribution for RL was accepted by NeurIPS 2025 (Oral). See you in San Diego!
  
  

Selected Publications

In submission

Data-regularized Reinforcement Learning for Diffusion Models at Scale

Haotian Ye, Kaiwen Zheng, Jiashu Xu, Puheng Li, Huayu Chen, Jiaqi Han, Sheng Liu, Qinsheng Zhang, Hanzi Mao, Zekun Hao, Prithvijit Chattopadhyay, Dinghao Yang, Liang Feng, Maosheng Liao, Junjie Bai, Ming-Yu Liu, James Zou, Stefano Ermon

[Paper] [Website] [Code] [Twitter]

In submission

Can Language Models Discover Scaling Laws?

Haowei Lin*, Haotian Ye*, Quzhe Huang, Wenzheng Feng, Yujun Li, Xiangyu Wang, Hubert Lim, Zhengrui Li, Jianzhu Ma, Yitao Liang, James Zou

[Paper] [Website] [Code] [Blog]

In submission

InfoTok: Adaptive Discrete Video Tokenizer via Information-Theoretic Compression

Haotian Ye*, Qiyuan He*, Jiaqi Han, Puheng Li, Jiaojiao Fan, Zekun Hao, Fitsum Reda, Yogesh Balaji, Huayu Chen, Sheng Liu, Angela Yao, James Zou, Stefano Ermon, Haoxiang Wang, Ming-Yu Liu

[Paper] [Code]

NeurIPS 2025 Oral

A Snapshot of Influence: A Local Data Attribution Framework for Online Reinforcement Learning

Yuzheng Hu, Fan Wu, Haotian Ye, David Forsyth, James Zou, Nan Jiang, Jiaqi W. Ma, Han Zhao

[Paper] [Code]

AISTATS 2025

Efficient and Asymptotically Unbiased Constrained Decoding for Large Language Models

Haotian Ye, Himanshu Jain, Chong You, Ananda Theertha Suresh, Haowei Lin, James Zou, Felix Yu

[Paper] [Code]

NeurIPS 2024 Spotlight

TFG: Unified Training-Free Guidance for Diffusion Models

Haotian Ye*, Haowei Lin*, Jiaqi Han*, Minkai Xu, Sheng Liu, Yitao Liang, Jianzhu Ma, James Zou, Stefano Ermon

[Paper] [Code] [Twitter]

Nature Machine Intelligence

A computational framework for neural network-based variational Monte Carlo with Forward Laplacian

Ruichen Li*, Haotian Ye*, Du Jiang, Xuelan Wen, Chuwei Wang, Zhe Li, Xiang Li, Di He, Ji Chen, Weiluo Ren, Liwei Wang

[Paper] [Code]

NeurIPS 2023 Oral

Towards Revealing the Mystery behind Chain of Thought: a Theoretical Perspective

Guhao Feng*, Bohang Zhang*, Yuntian Gu*, Haotian Ye*, Di He, Liwei Wang

[Paper] [Video] [Slides]

ICML 2023 Oral

On the Power of Pre-training for Generalization in RL: Provable Benefits and Hardness

Haotian Ye*, Xiaoyu Chen*, Liwei Wang, Simon Shaolei Du

[Paper]

ICLR 2023

Discovering Latent Knowledge in Language Models Without Supervision

Collin Burns*, Haotian Ye*, Dan Klein, Jacob Steinhardt

[Paper]

Invited Talks

• NVIDIA Research, Oct. 2025
  Data-regularized Reinforcement Learning for Diffusion Models at Scale

• NVIDIA Research, July 2025
  InfoTok: Adaptive Discrete Video Tokenizer via Information-Theoretic Compression

• Google Research, March 2025
  Efficient and Asymptotically Unbiased Constrained Decoding for Large Language Models

• The Hong Kong University of Science and Technology, Dec. 2023
  Towards Revealing the Mystery behind Chain of Thought: A Theoretical Perspective

• Google Brain, May 2023
  Towards Revealing the Mystery behind Chain of Thought: A Theoretical Perspective

Professional Services

• Reviewer: NeurIPS (2022-2025), ICLR (2024-2026), ICML (2024-2025), AISTATS (2023, 2025, Top Reviewer), ICCV (2025), EMNLP (2022)

Selected Awards

• Weiming Scholar of Peking University (1%), 2023
• Person of the Year of Peking University (10 people/year), 2021
• May 4 scholarship (1%, Rank 1), 2021
• National scholarship (1%, Rank 2), 2019
• Leo Koguan scholarship (1%), 2020
• Merit student pacesetter (2%), 2019
• Chinese Mathematical Olympiad (First Prize, Rank 7 in China), 2017

Miscellaneous

I spend most of my free time on photography, scuba diving (check the avatar!), and soccer (Palo Alto Adult Soccer League)!