Accelerating protein engineering with fitness landscape modelling and reinforcement learning, Nature Machine Intelligence 2025
Chemist-aligned retrosynthesis by ensembling diverse inductive bias models, arXiv 2025
NatureLM: Deciphering the Language of Nature for Scientific Discovery, arXiv 2025
HybriDNA: A Hybrid Transformer-Mamba2 Long-Range DNA Language Model, ICLR 2025 MLGenX Workshop
3DMolFormer: A Dual-channel Framework for Structure-based Drug Discovery, ICLR 2025
Token-level Direct Preference Optimization, ICML 2024
Connecting Large Language Models with Evolutionary Algorithms Yields Powerful Prompt Optimizers, ICLR 2024
Re-evaluating Retrosynthesis Algorithms with Syntheseus, Faraday Discussions 2024
De novo Drug Design using Reinforcement Learning with Multiple GPT Agents, NeurIPS 2023
QuinNet: Efficiently Incorporating Quintuple Interactions into Geometric Deep Learning Force Fields, NeurIPS 2023
Retrosynthetic Planning with Dual Value Networks
Research Experience
During his PhD studies and his first year as a researcher at Microsoft Research Asia, he gained extensive experiences in deep RL research and its applications in video games (e.g., Project Suphx: The World's Best Mahjong AI, and Project Mariana: Automated Game Testing with Xbox Studios). Now, he is a Senior Researcher at Microsoft Research AI for Science based in Cambridge, UK.
Education
He completed his Ph.D. from the University of Science and Technology of China (USTC) through a joint program with Microsoft Research Asia (2016-2021), under the supervision of Tie-Yan Liu and Nenghai Yu.
Background
His research interests include reinforcement learning (RL), large language models (LLMs), and AI for Scientific Discovery. Currently, he works on developing LLMs and RL agents to advance chemistry and drug discovery.