Research
My research focuses on the optimization and mathematical foundations of modern machine learning, especially large-scale overparameterized models. I study how loss landscapes, training dynamics, and optimizer design shape the behavior of these models.
A central question motivating my research is: Why do simple algorithms and ideas often work surprisingly well in practice, and what fundamental principles are behind it?
Papers
Showing recent and selected papers.
* denotes equal contribution, (α-β order) denotes alphabetical ordering
Service
Organizer
- IFDS Seminar Co-organizer: Fall 2025 (with Libin Zhu), Winter 2026, and Spring 2026 (with Weihang Xu).
Teaching
- CPS590.04 Machine Learning Algorithms, Spring 2021 @ Duke, TA.
- CPS330 Design and Analysis of Algorithms, Fall 2020 @ Duke, TA.
- CPS330 Design and Analysis of Algorithms, Spring 2020 @ Duke, TA.
Reviewing
- Reviewer for ICML, ICLR, NeurIPS, COLT, STOC, AAAI, AISTATS, CVPR, JMLR, Mathematical Programming, TPAMI, JASA.