About Me

Contact: xr021@ie.cuhk.edu.hk / ruichenxu@link.cuhk.edu.hk / rcxu642@gmail.com

I am a PhD candidate at The Chinese University of Hong Kong. My recent research focuses on understanding the theoretical foundations of how LLMs and neural networks learn (reasoning, knowledge mechanisms, etc.). I’m also broadly interested in mechanistic interpretability, AI safety, and privacy.

I am currently seeking research positions. If you are interested in collaboration or have an open position, please feel free to reach out!

Selected Publications

Filtering with Self-Attention and Storing with MLP: One-Layer Transformers Can Provably Acquire and Extract Knowledge
Ruichen Xu, Kexin Chen

Feature Resemblance: Towards a Theoretical Understanding of Analogical Reasoning in Transformers
Ruichen Xu, Wenjing Yan, Ying-Jun Angela Zhang
International Conference on Machine Learning, 2026.

Rethinking Benign Overfitting in Two-layer Neural Networks
Ruichen Xu, Kexin Chen
International Conference on Machine Learning, 2025.

Services

Reviewer of ICDCS 2024, AISTATS 2026, ICML 2026, NeurIPS 2026