Hengzhi He

I am a third-year PhD candidate at UCLA, working with Professor Guang Cheng. My research interests lie in applied probability, theoretical statistics, and theoretical computer science.

Email: hengzhihe@g.ucla.edu

Google Scholar

Current Focus

• Model Collapse: Understanding the benefits and risks of training models using synthetic data, and developing more efficient and reliable algorithms to leverage synthetic data.
• Statistical Watermarking: Designing watermarks to trace the origins of generated content, and developing more credible methods for evaluating watermarking techniques.

Selected Manuscripts

Watermarking Generative Tabular Data
Hengzhi He*, Peiyu Yu*, Junpeng Ren, Ying Nian Wu, Guang Cheng
Under Major Revision at Journal of the American Statistical Association (JASA), 2024
[arXiv]

Golden Ratio Weighting Prevents Model Collapse
Hengzhi He*, Shirong Xu*, Guang Cheng
Under Major Revision at Journal of the Royal Statistical Society Series B (JRSSB), 2025
[arXiv]

"Noisier" Noise Contrastive Estimation is (Almost) Maximum Likelihood Estimation
Peiyu Yu*, Dinghuai Zhang*, Hengzhi He*, Xiaojian Ma, Sirui Xie, Ruiyao Miao, Yifan Lu, Yasi Zhang, Deqian Kong, Ruiqi Gao, Jianwen Xie, Guang Cheng, Ying Nian Wu
Accepted at International Conference on Learning Representations (ICLR), 2026
[OpenReview]

Optimal Watermark Generation under Type I and Type II Errors
Hengzhi He, Shirong Xu, Alexander Nemecek, Jiping Li, Erman Ayday, Guang Cheng
Under R&R at Biometrika, 2026
[arXiv]

A Probabilistic Perspective on Model Collapse
Shirong Xu*, Hengzhi He*, Guang Cheng
Submitted, 2025
[arXiv]

* denotes equal contribution. For a full list, please refer to my Google Scholar page.

Education

• Department of Statistics and Data Science, University of California, Los Angeles (UCLA), 2023 – Present
  PhD Student, Advisor: Prof. Guang Cheng
• School of Mathematical Sciences, Peking University, 2019 – 2023
  Undergraduate Student