Research Interests

I work on random matrix theory and transformer mechanistic interpretability, recently started exploring stochastic differential equations, and am still searching for the theory that unifies them with jazz.

Publications & Manuscripts

Mechanistic Interpretation

Task Vector Geometry Underlies Dual Modes of Task Inference in Transformers
H. Yan, H. Yang and Y. Zhong
Preprint, 2026

Bandit and Reinforcement Learning

Learning with Incomplete Context: Linear Contextual Bandits with Pretrained Imputation
H. Yan H. Zhang and Y. Guo
AISTATS, 2026

Entrywise Low-rank Matrix Estimation

Estimating Multiple Weighted Networks with Node-Sparse Differences and Shared Low-Rank Structure
H. Yan and K. Levin
Preprint, 2025
Improved dependence on coherence in eigenvector and eigenvalue estimation error bounds
H. Yan and K. Levin
AISTATS 2025
Coherence-free Entrywise Estimation of Eigenvectors in Low-rank Signal-plus-noise Matrix Models
H. Yan and K. Levin
NeurIPS 2024
Minimax rates for latent position estimation in the generalized random dot product graph
H. Yan and K. Levin
Preprint, 2023