Zhaoyang Shi

Zhaoyang Shi

Assistant Professor of Center for Applied Mathematics, Fudan University

I am an Assistant Professor of the Center for Applied Mathematics at Fudan University, with a joint appointment in the School of Mathematical Sciences. From 2024 to 2026, I was a Postdoctoral Fellow in the Department of Statistics at Harvard University, under the supervision of Prof. Zheng (Tracy) Ke. I received my Ph.D. in Statistics from UC Davis, where I was co-advised by Prof. Krishna Balasubramanian and Prof. Wolfgang Polonik. Before UC Davis, I received my B.S. in Mathematics from Fudan University, where I was advised by Prof. Lei Shi.

My research lies at the intersection of statistics, machine learning, and artificial intelligence, with an emphasis on developing theory and methods for reliable statistical inference, understanding modern AI models, and solving complex scientific and data-driven problems.

Research

Statistical Inference & Learning
Nonparametric and high-dimensional statistics, network analysis, Gaussian and bootstrap approximation, and uncertainty quantification.
Foundations of Generative AI
Associative memory, energy-based models, diffusion models, and flow-based generative models.
Data Science & Decision Making
AI and statistical methods for scientific discovery, complex systems, interdisciplinary data, and data-driven decision-making.
Collaborations: I am always open to collaborations and new research directions. Feel free to reach out if you are interested in working together.

Recent News

New Position I joined the Center for Applied Mathematics at Fudan University, with a joint appointment in the School of Mathematical Sciences, as a tenure-track Assistant Professor on Sep 3 2026.
CMStatistics 2026 Invited talk: Mixed Membership Amid Dynamic Networks, the 20th International Joint Conference on Computational and Financial Econometrics (CFE) and Computational and Methodological Statistics (CMStatistics), HTW Berlin, Germany, Dec 12-Dec 14, 2026.
NEW Our work Population-Level Generative Modeling for Ranking Data is now available on arXiv. We introduce latent preference simplex embedding with flow matching (LPSE-FM) for accurate synthetic ranking generation and interpretable modeling of preference heterogeneity.
ICLR 2026 I will co-organize the workshop New Frontiers in Associative Memory.
AoS Gaussian and Bootstrap Approximation for Matching-based Average Treatment Effect Estimators has been accepted by Annals of Statistics. Using ATE estimation in causal inference as a motivating example, we develop a general framework for non-asymptotic statistical inference through local geometry and stabilization.
NeurIPS Dense Associative Memory with Epanechnikov Energy was accepted as a Spotlight (top 3%). We propose the log-sum-ReLU (LSR) energy, inspired by the optimal kernel in kernel density estimation, to address the memorization-generation trade-off in dense associative memories.
SIAM UQ26 Invited talk: From Smooth to Nonsmooth: Minimax Optimal Regression with Laplacian Eigenmaps, at the minisymposium Probabilistic Manifold Learning and Deep Embeddings for Uncertainty Quantification, March 2026.
JSM 2025 Invited talk: Smooth Dynamic Network Analysis, at JSM 2025, August 2025.
APS 2025 Talk: On the Nonasymptotic Statistical Inferences via Stabilization Theory of Gaussian Approximation Bounds, at the 22nd INFORMS Applied Probability Society Conference, Georgia Institute of Technology, June 30–July 3, 2025.
2024 I joined the Department of Statistics at Harvard University as a Postdoctoral Fellow on September 1, 2024.