Jiaxuan Zou

School of Mathematics and Statistics Xi'an Jiaotong University

Hi, I am Jiaxuan Zou (邹嘉轩).

I am an undergraduate student in Mathematics and Statistics at Xi’an Jiaotong University and currently a research intern in the ByteDance Seed Pre-training team. Previously, I was a research intern at the Gaoling School of Artificial Intelligence, Renmin University of China, advised by Prof. Yong Liu.

My research focuses on mechanistic interpretability, deep learning theory, optimization, and scaling laws. I am especially interested in turning empirical training phenomena into first-principles mechanisms: why models train, where behaviors emerge, and when scaling or optimization rules break.

I also work as an AI Technical Consultant for a Tsinghua-affiliated AI startup.

I write research notes on my blog and welcome conversations on these topics. I am interested in future research opportunities in LLM pre-training and AI theory, as well as PhD positions for Fall 2028. More background: English / 中文.

Research Interests

Mechanistic interpretability of LLMs
Training dynamics of finite-width networks
Optimizer design for LLM pre-training
Scaling laws and their failure modes
Deep learning theory and optimization

News

Jul 01, 2026	I joined the ByteDance Seed Pre-training team as a research intern.
May 25, 2026	Our Nora optimizer has been included in the ScalingOpt optimizer library.
Apr 18, 2026	I joined the ScalingOpt project as a co-maintainer, working on optimizer design for large language model training.

Latest Posts

Jul 04, 2026	本科两年，我最深的感悟：认知复利
Jun 20, 2026	DASF：一种闭环的 batch size schedule-free 方法
Jun 16, 2026	为什么 LLM pretrain 过程中途要把 batch size 翻倍
Jun 03, 2026	不要只学习 19 世纪的西方：文明中心论、世界主义与青年领袖的公共责任
May 24, 2026	重听杨植麟：Bet on Scaling、第一性原理和长期主义
May 15, 2026	μP Map
Apr 14, 2026	在 LLM 语境下，梯度里的噪声会如何影响 training dynamics？
Mar 23, 2026	并行性与表达能力的权衡：从 $AC^0$/$TC^0$ 到 Linear Attention 的理论边界

Selected Publications

Kaczmarz Linear Attention

Jiaxuan Zou, Ruifeng Ren, and Yong Liu

2026

Bib PDF

@misc{zou2026kaczmarzlinearattention,
  title = {Kaczmarz Linear Attention},
  author = {Zou, Jiaxuan and Ren, Ruifeng and Liu, Yong},
  year = {2026},
  journal = {Under Review},
  archiveprefix = {arXiv},
  primaryclass = {cs.LG},
  url = {https://arxiv.org/abs/2605.08587},
}

Nora: Normalized Orthogonal Row Alignment for Scalable Matrix Optimizer

Jinghui Yuan, Jiaxuan Zou, Shuo Wang, and 2 more authors

2026

Bib PDF

@misc{yuan2026noranormalizedorthogonalrow,
  title = {Nora: Normalized Orthogonal Row Alignment for Scalable Matrix Optimizer},
  author = {Yuan, Jinghui and Zou, Jiaxuan and Wang, Shuo and Liu, Yong and Nie, Feiping},
  year = {2026},
  journal = {Under Review},
  archiveprefix = {arXiv},
  primaryclass = {cs.LG},
  url = {https://arxiv.org/abs/2605.03769},
}

Effective Frontiers: A Unification of Neural Scaling Laws

Jiaxuan Zou, Zixuan Gong, Ye Su, and 2 more authors

Under Review, 2026

Bib PDF

@article{zou2025effective,
  title = {Effective Frontiers: A Unification of Neural Scaling Laws},
  author = {Zou, Jiaxuan and Gong, Zixuan and Su, Ye and Tang, Huayi and Liu, Yong},
  year = {2026},
  journal = {Under Review},
  archiveprefix = {arXiv},
  primaryclass = {cs.LG},
}

Capabilities and Fundamental Limits of Latent Chain-of-Thought

Jiaxuan Zou, Yaozhong Xiong, and Yong Liu

Under Review, 2026

Bib PDF

@article{zou2025capabilities,
  title = {Capabilities and Fundamental Limits of Latent Chain-of-Thought},
  author = {Zou, Jiaxuan and Xiong, Yaozhong and Liu, Yong},
  year = {2026},
  journal = {Under Review},
  archiveprefix = {arXiv},
  primaryclass = {cs.AI},
}

Statistical MIA: Rethinking Membership Inference Attack for Reliable Unlearning Auditing

Jialong Sun, Zeming Wei, Jiaxuan Zou, and 5 more authors

Under Review, 2026

Bib PDF

@article{sun2025statistical,
  title = {Statistical MIA: Rethinking Membership Inference Attack for Reliable Unlearning Auditing},
  author = {Sun, Jialong and Wei, Zeming and Zou, Jiaxuan and Gong, Jiacheng and Wang, Guanheng and Dong, Chengyang and Li, Jialong and Liu, Bo},
  year = {2026},
  journal = {Under Review},
  archiveprefix = {arXiv},
  primaryclass = {cs.LG},
}

FreIE: Low-Frequency Spectral Bias in Neural Networks for Time-Series Tasks

Jialong Sun, Xinpeng Ling, Jiaxuan Zou, and 2 more authors

In IEEE ICDM, 2025

Bib PDF Video

@inproceedings{sun2025freielowfrequencyspectralbias,
  title = {FreIE: Low-Frequency Spectral Bias in Neural Networks for Time-Series Tasks},
  author = {Sun, Jialong and Ling, Xinpeng and Zou, Jiaxuan and Kang, Jiawen and Zhang, Kejia},
  year = {2025},
  archiveprefix = {arXiv},
  primaryclass = {cs.LG},
  url = {https://arxiv.org/abs/2510.25800},
  booktitle = {IEEE ICDM}
}