Xinyan Hu

Hi! I am a CS PhD student at UC Berkeley and a member of the Berkeley AI Research (BAIR) Lab, advised by Michael I. Jordan and Jacob Steinhardt. I'm also an Anthropic Fellow working on security of LLM agents, in particular, whether backdoors planted during pretraining survive post-training in agentic models. More broadly, my research aims to improve model reliability by studying training dynamics and inference procedures, drawing on mechanistic interpretability, algorithmic game theory, and machine learning theory.

Before coming to Berkeley, I received my B.S. in CS at Peking University, where I was fortunate to work with Xiaotie Deng on algorithmic game theory, auction theory and online learning. I also had the pleasure of working with Zhiwei Steven Wu (CMU) and Aleksandrs Slivkins (MSR) on multi-armed bandits and aligning online learning algorithms with human incentives.

Selected Papers

Understanding In-context Learning of Addition via Activation Subspaces
Xinyan Hu, Kayo Yin, Michael I Jordan, Jacob Steinhardt*, Lijie Chen*
First Workshop on the Interplay of Model Behavior and Model Internals @ COLM 2025
[arXiv][twitter]
Incentivizing High-Quality Content in Online Recommender Systems

Xinyan Hu*, Meena Jagadeesan*, Michael I. Jordan, Jacob Steinhardt

FORC 2026 (🏆 Best Student Paper Award)

[arXiv]

Incentivizing Combinatorial Bandit Exploration

Xinyan Hu, Dung Daniel Ngo, Aleksandrs Slivkins, Zhiwei Steven Wu (in alphabetical order)

NeurIPS 2022

[arXiv]

Nash Convergence of Mean-Based Learning Algorithms in First Price Auctions

Xiaotie Deng, Xinyan Hu, Tao Lin, Weiqiang Zheng (in alphabetical order)

The Web Conference (WWW) 2022

[arXiv]

Selected Awards & Fellowship

Anthropic AI Safety Research Fellow, 2025
Leo KoGuan scholarship, 2021
Yang Fuqing & Wang Yangyuan Academician Scholarship, 2020
Chinese Girls' Mathematical Olympiad (CGMO), Gold Medal, 2017

Miscellaneous

I have a deep passion for the outdoors—hiking and biking to immerse myself in nature—and love photographing to capture the fleeting beauty in everyday life and nature. In my spare time, I enjoy listening to music and playing various sports, particularly tennis. Novak Djokovic has been my idol and a significant inspiration in my personal growth since I was 11 years old. In a memorable adventure, I was fortunate enough to meet him in person, securing his autograph and capturing a selfie with him at BNP Paribas Open in Indian Wells in 2024!

Page updated

Google Sites

Report abuse