Most Significant 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
(*Equal Contribution)
International Conference on Learning Representations (ICLR) 2022, 26 April 2022. (Oral)
🏆 ICLR 2022 Outstanding Paper Award
*Equal Contribution. Sometimes non-alphabetical to balance visibility in citations.
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 (2025).
💬 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.
X.Y. Han 