UC Berkeley / Operations Research
Stochastic nonconvex optimization
and machine learning.
I am a PhD student in Operations Research at the University of California, Berkeley, co-advised by Ying Cui and Huiwen Jia. My research explores the mathematical foundations of optimization and their connections to learning.
I study optimization algorithms in the presence of randomness, motivated by the methods used in machine learning. I am interested in how noise shapes their behavior, what can be guaranteed, and how theoretical insights can inform algorithm design.
I am also interested in decision-dependent optimization, where decisions influence the data available for learning. With a limited budget, how should we balance improving a decision with collecting new data? This perspective brings learning and optimization into the same iterative process.
A library of oracle models and complexity bounds for stochastic optimization under heavy-tailed noise. Formal statements make the assumptions and the information available to an algorithm explicit.
heavy-tailed-noise-leanA hypergraph-based framework for large-scale quadratically constrained quadratic programs, combining learned solution prediction with neighborhood optimization to improve candidate solutions.
Read the paperA benchmark and toolkit for curriculum learning, comparing training curricula across vision, language, and graph tasks through both learning performance and computational complexity.
Read the paperBefore Berkeley, I studied mathematics, physics, and industrial engineering at Tsinghua University's Weiyang College. In Fall 2023, I spent a semester at the University of Toronto, working with Chi-Guhn Lee. I received an M.S. in Engineering from Berkeley in May 2026.
Outside research, I enjoy tennis and skiing.
| Sep 27, 2026 | Released heavy-tailed-noise-lean, a Lean 4 formalization of lower bounds for strict-K=1 stochastic optimization under heavy-tailed noise. |
|---|
For research questions and collaborations, please get in touch.
xiongzhixiao@berkeley.edu