Yixuan Yang (杨怡轩)

Ph.D. Candidate

School of Mathematical Sciences
Zhejiang University

Office: 506, Hainayuan Building #2
Email: yixuanyang [at] zju [dot] edu [dot] cn

About me

I am a fifth-year Ph.D. candidate in the School of Mathematical Sciences at Zhejiang University, advised by Prof. Song Li. Prior to that, I received my B.S. degree in Pure and Applied Mathematics from Lanzhou University in 2022.

Please feel free to contact me if you have any questions or would like to discuss my research.

News

[2026.11.03] Our preprint Theoretical Analysis of Muon on Transformers for Next-Token Prediction is available on arXiv.
[2026.05.26] Our preprint Convergence of Spectral Descent for Non-smooth Optimization is available on arXiv.
[2026.05.18] We uploaded math-beamer-slides-skill.
[2026.05.03] Our preprint How Label Imbalance Shapes Geometry: A General Spectral Analysis of Multi-Label Neural Collapse is available on arXiv.
[2025.12.01] Our paper The implicit regularization of gradient flow on separable datasets in ReLU networks appeared in Neurocomputing.

Research Interests

My research focuses on the mathematical foundations of data science, with an emphasis on deep learning theory and non-smooth optimization. I am also interested in the theory, algorithms, and applications of Transformers and large language models.

Data Science Problems: low-rank matrix recovery, phase retrieval, and compressed sensing.
Deep Learning Theory: optimization, generalization, and explanation of neural collapse phenomena.
Optimization: convergence and implicit bias of algorithms.

Recently, I have been exploring how AI can accelerate mathematical research, particularly through the development of relevant skills and research agents. I welcome discussions and potential collaborations with others working in this direction.

Academic Service

ICLR Reviewer

Publications / Preprints

* denotes equal contribution

Theoretical Analysis of Muon on Transformers for Next-Token Prediction. Shengwei Wu^*, Yixuan Yang^*, Gao Huang^*, Song Li arXiv:2611.04465, 2026.
Convergence of Spectral Descent for Non-smooth Optimization. Yixuan Yang, Yuqing He, Song Li arXiv:2605.26977, 2026.
How Label Imbalance Shapes Geometry: A General Spectral Analysis of Multi-Label Neural Collapse. Xiaoxuan Ma, Yixuan Yang, Song Li, Xiangyun Hui arXiv:2605.01897, 2026.
The implicit regularization of gradient flow on separable datasets in ReLU networks. Xiangyun Hui, Xiaoxuan Ma, Yixuan Yang, Song Li Neurocomputing, 2025.