Research

My research is centered around machine learning with particular interests in reinforcement learning. My research goal is to develop and understand the foundation of decision making processes for complicated tasks such as LLM, agentic systems and structural design. I’m also interested in developing AI agents for scientific (chemistry, physics) tasks. In particular: I’m actively working on

Theoretical foundation of RL

I am interested in understanding some center topics in RL that would inspire the empirical design of the RL algorithms. In particular, we are recently working on:

• Unsupervised exploration in reinforcement learning (ICML24)
• Misspecification-robust reinforcement learning (NeurIPS24)

Empirical of RL with foundation models

The raise of foundation models provide potentials for improve the performance of RL algorithms, and the insights from RL are also widely used in foundation models. In particular, some of my recent works cover

• Diffusion models / Flow-based Generative Models for RL (ICLR25)
• Robust self-improvement for LLMs (ICLR25)

Broad Scientific Tasks

I’m working on extending the reinforcement learning (and general machine learning) to help automate many self-driving lab and design the exploration of the experiments. Some of the special interests covers

• De-novo drug design (in collaboration with Nvidia and Astrazeneca)
• Machine learning / Reinforcement learning for catalyst discovery (in collaboration with Prof. Chong Liu at UCLA)

I’m actively seeking for more interdisciplinary collaborations because different science tasks require specific design of the ML algorithm.

Acknowledgement

We are gratefully appreciate the support from Google Cloud and Nvidia