Zhiyuan Li

Zhiyuan Li is a tenure-track assistant professor at the Toyota Technological Institute at Chicago (TTIC), affiliated faculty in Computer Science at the University of Chicago, and visiting faculty at Google Research. He previously served as a postdoctoral fellow in the Computer Science Department at Stanford University, working with Tengyu Ma. Li obtained his PhD in Computer Science from Princeton University in 2022, advised by Sanjeev Arora, an MS from Princeton in 2019, and a BS from Tsinghua University's Yao Class in 2017. His career trajectory reflects a commitment to advancing theoretical machine learning.

Li's research focuses on deep learning theory, including optimization algorithms' implicit biases, generalization in overparameterized networks, sharpness-aware methods, and large language models' reasoning via chain-of-thought. Key publications include 'Fine-Grained Analysis of Optimization and Generalization for Overparameterized Two-Layer Neural Networks' (ICML 2019; 1,345 citations), 'On Exact Computation with an Infinitely Wide Neural Net' (NeurIPS 2019; 1,305 citations), 'Why Are Convolutional Nets More Sample-Efficient than Fully-Connected Nets?' (ICLR 2021, Oral), 'What Happens after SGD Reaches Zero Loss? – A Mathematical Framework' (ICLR 2022, Spotlight), 'How Does Sharpness-Aware Minimization Minimize Sharpness?' (ICLR 2023), and 'Chain of Thought Empowers Transformers to Solve Inherently Serial Problems' (ICLR 2024). With over 7,400 citations, his work shapes modern AI training paradigms. Awards include the Microsoft Research PhD Fellowship (2020), William G. Bowen Merit Fellowship (2017), Outstanding NeurIPS Reviewer Award (2021), and multiple ICLR Spotlights/Oral presentations. Li contributes as NeurIPS 2023 Area Chair and reviewer for JMLR, ICML, ICLR, and others.