李虓 Xiao Li

About Me

I am currently a Postdoctoral Fellow at the University of Hong Kong, working with Prof. Yi Ma. Previously, I received my Ph.D. in EECS from the University of Michigan, where I was fortunate to be advised by Prof. Qing Qu.
My research interests lie in representation learning, focusing particularly on the identification and effective utilization of low-dimensional structures and dynamics of deep networks. Besides, I’m also interested in robust deep learning, and Super-resolution microspcopy.

[May 2026] Paper on evaluating and monitoring diffusion model training via representation space geometry (project page) accepted at ICML 2026.
[Dec. 2025] Paper on how balanced representation space drives generalization in diffusion models (project page) accepted at ICLR 2026.
[Sep. 2025] Paper on diffusion-based representation learning dynamics accepted at NeurIPS 2025.
[Aug. 2025] Paper on layerwise representation dynamics in deep networks accepted at JMLR.
[May 2024] Paper on transfer learning and neural collapse accepted at TMLR; paper on multi-label learning and neural collapse accepted at ICML 2024.