Haolin Liu

Hi! I am Haolin Liu, a third-year PhD student at the University of Virginia, where I am fortunate to be advised by Prof. Chen-Yu Wei. Prior to this, I received my bachelor’s degree in Computer Science from ShanghaiTech University, where I studied chemistry for 1.5 years before transitioning to computer science for 2.5 years.

I am interested in developing principled and practical algorithms for Reinforcement Learning (RL), and understanding the training dynamic of these algorithms. Recently, I mainly focus on RL theory, RL for LLM reasoning and agents.

• On the theoretical side, I study unified principles for RL algorithm design and seek to characterize the minimal structure required for sample-efficient learning. My recent works ([1], [2]) propose the most unified RL-theory frameworks to date, capable of handling both model-based and model-free RL in stationary and non-stationary environments.
• On the practical side, I develop scalable RL pipelines for self-improving LLM agents with advanced reasoning and continual learning capabilities. My prior work has centered on new RL methods with fine-grained supervision and enhanced exploration mechanisms for LLM reasoning. I am currently building new environments and RL algorithms that enable agents to tackle long-horizon, complex tasks.

Currently, I am an intern at Bytedance Seed-LLM in San Jose, where I work on RL for multi-turn interactive agents. Previously, I was an intern at at Tencent AI Lab in Seattle.