My research lies in the interface of machine learning/non-parametric statistics, applied probability and numerical computation, with a primary focus on the algorithmic and theoretical foundations for more “modern” settings such as Deep Learnig, Generative AI and AI4Science.


Scientific Machine Learning/Machine Learning for Science

Validated Scientific Machine Learning via Debiasing

Coming soon, here’s our first paper on theory

Scaling Law of Scientific Machine Learning

How large the sample size and how much computational power are needed to reach a prescribed performance level for a physic problem?

Encoding Physics Information into a Model

Robust Machine Learning

Experiment Design

Balanced Experiment Design: In this arxiv paper, we surprisingly connect balanced experiment design for panel data with phase retrieval problem from the inverse problem community. This connection enables us to generate the design via a generalized power method. For economists, our paper discovered that best experiment = smallest (nonlinear) eigenvector for the smallest eigenvector can cancel most linear effects.

Other Projects