Assistant Professor
Operations Management & Applied AI
John E. Jeuck Faculty Fellow
Booth School of Business
University of Chicago

Affiliations:
Center for Applied AI
Tolan Center for Healthcare

Education

• BSE in Operations Research and Financial Engineering, Princeton
• PhD in Operations Research and Information Engineering, Cornell
  Supervisor: Adrian S. Lewis
• Post-doc & MS in Statistics, Stanford
  Supervisor: David L. Donoho

NAME: Preferred name is X.Y.
CITIZENSHIP: United States 🇺🇸
EMAIL: XY (dot) Han (at) chicagobooth (dot) edu

Curriculum Vitae

Most Recognized Work:
Discovered the neural collapse phenomenon in AI training [PNAS 2020; ICLR 2022].

---

Awards

🏆 ICLR 2022 Outstanding Paper Award

🏆 ICCOPT 2022 Best Paper Prize for Young Researchers (Finalist)

---

Publication Highlights

Most Influential Papers

Prevalence of Neural Collapse During the Terminal Phase of Deep Learning Training
Vardan Papyan*, X.Y. Han*, and David L. Donoho
Proceedings of the National Academy of Sciences (PNAS), 117.40 (2020): 24652-24663.
🌟 Discovered neural collapse, now a widely studied phenomenon in AI training.

Neural Collapse Under MSE Loss: Proximity to and Dynamics on the Central Path
X.Y. Han*, Vardan Papyan*, and David L. Donoho
International Conference on Learning Representations (ICLR) 2022, 26 April 2022. (Oral)
🏆ICLR 2022 Outstanding Paper Award

Recent Highlight(s)

A Theoretical Framework for Auxiliary-Loss-Free Load Balancing of Sparse Mixture-of-Experts in Large-Scale AI Models
X.Y. Han* and Yuan Zhong*
arXiv preprint arXiv:2512.03915 (2026).
💬 What can operations contribute to AI? This paper is my first answer. s-MoE load-balancing procedures are examples of ML heuristics that work very well in practice but we don’t have much rigorous understanding for why they would. In this paper, Yuan and I use mathematical tools from OR/OM to build a theoretical framework for analyzing why DeepSeek’s ALF-LB procedure is effective at s-MoE load-balancing.

*Equal Contribution. Sometimes non-alphabetical to balance visibility in citations.

Research Interests

Deconstructing the conventional wisdoms of AI.

Modern AI is built on a foundation of conventional wisdoms — architectures, optimizers, MLOps flows etc — discovered through countless feats of trial-and-error and engineering. Collectively, they enable AI to now perform near human-level decision-making, but we can’t precisely pinpoint why. My research deconstructs these successes to uncover the operational “first principles” of AI behavior.

Artificial Intelligence: Uncovering the “first principles” of AI by running large-scale experiments on used-in-practice AI architectures. Formulating mathematically-grounded models of their behavior from those experiments. (Prev. Work: Neural Collapse, MoE Load-Balancing)
Optimization: Mathematically distilling the fundamental structures present in nonsmooth, nonconvex, and nonlinear optimization problems. Using this knowledge to build faster and simpler optimization algorithms from the ground up. (Prev. Work: Survey Descent)
Applications: Translating this knowledge into productivity-enhancing tools. Focusing on solutions that amplify professional insights. Solving critical operational problems for collaborators in industry and civic institutions. (Past/Present Collabs: Frick Art Reference Library, Veolia North America LLC, Surgical Data Science Collective)

---

---