Tsinghua University · IIIS

Guanheng Chen

Hi, I’m Guanheng, a third-year undergraduate at the Institute for Interdisciplinary Information Sciences, Tsinghua University.

I’m interested in LLM post-training, particularly how language models can learn from experience and natural-language feedback.

I started my research journey at Microsoft Research Asia, and I am fortunate to be mentored by Dr. Li Dong and to work with Tianzhu Ye.

Email Curriculum vitae GitHub

Selected research

Figure 1 from Learning to Coach: accuracy improves and response length decreases over coaching iterations.

Learning to Coach for Experiential Learning

Guanheng Chen^*, Tianzhu Ye^*, Li Dong^*, Xun Wu, Shaohan Huang, Furu Wei

Train a coach to turn an actor’s past attempts into useful guidance. The actor stays frozen; the coach learns from its downstream performance, enabling iterative improvement and transferable experiential knowledge.

Paper PDF

Figure 1 from LLM-as-a-Coach: comparison of a judge's scalar reward with a coach's natural-language experiential knowledge.

LLM-as-a-Coach: Experiential Learning for Non-Verifiable Tasks

Tianzhu Ye^*, Li Dong^*, Guanheng Chen, He Zhu, Xun Wu, Shaohan Huang, Furu Wei

Learn from rich natural-language feedback on open-ended tasks. Experiential knowledge is internalized through on-policy context distillation, providing a richer learning signal than scalar rewards.

Paper PDF

* Equal contribution.

Experiences

Mar 2026 Present

Microsoft Research Asia

Research Intern

General Artificial Intelligence Group

Beijing, China

Aug 2024 Present

Tsinghua University

Third-year Undergraduate

Institute for Interdisciplinary Information Sciences

Beijing, China

Misc

Basketball

One of my favorite things to do is watch live NBA games on weekend mornings. Hoping to watch a Lakers game in Staples Center someday🏀.

Reading

Recently, I'm obsessed with science fiction. My favorite short story is With Her Eyes by Liu Cixin.

Games

I’ve been playing Hearthstone for ten years.