Yunzong Xu

I study the mathematical foundations of machine learning and sequential decision-making. The problems I think about usually contain some kind of friction: not enough data, not enough information, not enough time, not enough resources, or not enough freedom to try everything. Theory is a way of asking which frictions are reducible and which ones are fundamental.

Before joining UIUC in 2024, I was a postdoctoral researcher at Microsoft Research. Before that I completed my PhD at MIT and my undergraduate studies at Tsinghua University.

My work develops mathematical foundations for algorithms that learn from data and make decisions under uncertainty. Topics include online learning, deep learning, bandit algorithms, reinforcement learning, stochastic networks, information theory, and game theory, with applications to AI systems, communication networks, online marketplaces, data centers, and supply chains. I am interested in how algorithms acquire and use information over time, how geometric, statistical, and network structures shape what is achievable, and where the fundamental limits lie.

I am fortunate to work with the following students.

Besides, I’ve been working with teddy bears for as long as I can remember. You’ll be able to see some of them if you pass by my office.