I am Hugh Zheng, a Ph.D. candidate in Statistics at the University of Florida (UF), advised by Dr. Leo L. Duan. I publish as Yu Zheng.
My research builds probabilistic models for combinatorial and structured data, with a focus on the algorithms and proofs that make them work at scale. Recent projects include anti-correlation Gaussian data augmentation for sampling under high-dimensional constraints, integer-program-based likelihoods for combinatorial response data, and a 3D-convolutional variational autoencoder for resting-state fMRI. I work in Python, R, and C++, with PyTorch and CUDA for deep generative models and Rcpp for the C++ samplers. I prove the convergence theorems and write the code that runs them.
What draws me to this work is the interplay between mathematical structure and modern machine learning. The most interesting problems sit where probabilistic theory meets practical inference: how to sample efficiently when the parameter space has hard constraints, how to train deep generative models on medical imaging without overfitting, how to keep the optimization landscape tractable as the model grows.
Beyond research, I care about making the methods usable. My three publications each ship with reproducibility code, and I maintain open-source tools including latex2arxiv, a LaTeX-to-arXiv pipeline distributed as a PyPI package, a GitHub Action, and an MCP server. Writing the package usually teaches me what the paper actually claims.
Research Interests
- Bayesian asymptotics
- Model-based clustering
- Markov chain Monte Carlo (MCMC)
- Combinatorial and structured data analysis
- Nonparametric Bayesian methods
Education
Ph.D. in Statistics (Ongoing)
University of Florida
B.S. in Mathematics and Applied Mathematics, 2016-2020
University of Science and Technology of China
Work Experience
Quantitative Research Intern, Jun 2024 - Aug 2024
Susquehanna International Group (SIG)