Hanxuan Li 李瀚轩
 |
Hanxuan Li |
About me
I am a fourth-year undergraduate student (2022.9 – Present) in Turing Class, Chu Kochen Honors College, Zhejiang University. I am also an incoming PhD student at SphereLab, the Chinese University of Hong Kong, advised by Prof. Weiyang Liu. During my undergraduate studies, I have had the privilege of working with Prof. Yulun Zhang and Prof. Huan Wang.
My research vision is to understand the principles underlying large language model pretraining, with emphasis on optimization, architecture, and scaling laws. I am also interested in whether neural networks acquire structured internal models of tasks and environments and whether they can support causal representation and reasoning beyond surface correlation.
Pretraining principles
- Optimization. Algorithms and dynamics of LLM pretraining (e.g., stability, implicit bias, and efficiency).
- Architecture. Inductive biases in large-scale model designs, with emphasis on neural memory.
- Scaling laws. How performance and behavior scale with width, depth, and data, and how these axes interact.
News
Selected Papers
(* denotes equal contribution, † denotes corresponding author)
Pion: A Spectrum-Preserving Optimizer via Orthogonal Equivalence Transformation
|
For the full list, see Publications.