Haoqun Cao

Haoqun Cao 

Ph.D. Student, Department of Statistics
University of Wisconsin–Madison
Email: hcao65@wisc.edu
GitHub  |  LinkedIn  |  Google Scholar  |  CV

About Me

I am a third-year Ph.D. student in Statistics at the University of Wisconsin–Madison, advised by Prof. Tengyang Xie. I received my B.S. in Statistics from Renmin University of China in 2024, where I worked with Prof. Feng Zhou.

I am looking for research internship positions for Summer 2027. Please feel free to reach out!

Besides research, I enjoy soccer and swimming. I also spent four wonderful years in my college Chinese orchestra.

Research

I am interested in how to learn a policy that acts in a very large action space over a long horizon. This includes continuous control in the physical world, and also LLMs. In particular, during my Ph.D., I have worked on:

  • Action tokenization for long-horizon control

  • Reinforcement learning for LLM post-training

  • Long-horizon reasoning with LLMs

In the past, I worked on deep probabilistic models, including point processes and approximate inference. I lead my projects end to end, from problem formulation and theory to implementation, experiments and writing.

Publications

(* denotes equal contribution)

  • Anatomy of Off-Policy Policy Gradient: Importance Sampling, KL Regularization, and Baselines [arXiv]
    Haoqun Cao, Yurun Yuan, Tengyang Xie
    NeurIPS 2026

  • Understanding Behavior Cloning with Action Quantization [arXiv]
    Haoqun Cao, Tengyang Xie
    ICML 2026

  • Is Score Matching Suitable for Estimating Point Processes? [arXiv] [code]
    Haoqun Cao, Zizhuo Meng, Tianjun Ke, Feng Zhou
    NeurIPS 2024

  • Accelerating Convergence in Bayesian Few-Shot Classification [arXiv] [code]
    Tianjun Ke, Haoqun Cao, Feng Zhou
    ICML 2024

  • Revisiting Logistic-Softmax Likelihood in Bayesian Meta-Learning for Few-Shot Classification [arXiv] [code]
    Tianjun Ke*, Haoqun Cao*, Zenan Ling, Feng Zhou
    NeurIPS 2023

Preprints

  • Self-Improving Long-Horizon Reasoning by Learning Summaries from Local Continuations
    Haoqun Cao, Tengyang Xie
    Preprint, 2026 (draft available upon request)

  • Score Matching for Estimating Finite Point Processes [arXiv]
    Haoqun Cao, Yixuan Zhang, Feng Zhou
    Under review