Zonglin Yang

Zonglin Yang

Always aiming to make the next work better than the last.

About me

Hi, thanks for stopping by.

I am Zonglin Yang, a research scientist at MiroMind AI, where I work on LLMs for scientific discovery and large reasoning models.

Previously, I completed my Ph.D. at Nanyang Technological University supervised by Erik Cambria. I also worked closely with Soujanya Poria. Before NTU, I obtained my master’s degree from Cornell University, supervised by Claire Cardie and Xinya Du, and my bachelor’s degree at Huazhong University of Science and Technology, supervised by Xinggang Wang. I have interned at Microsoft Research in the NLC group mentored by Li Dong, and at MiroMind mentored by Lidong Bing. I was also a visiting student at Princeton University hosted by Mengdi Wang.

I try to focus on research questions that matter at a fundamental level, and to approach them with as much rigor and depth as I can. My goal is to make each new piece of research a little better than the last.

I am currently looking for interns to work on AI for Research. If you are interested, please send your CV to zonglin.yang@miromind.ai.

Research

My main research interest is LLMs for scientific discovery. Much of my work in this area centers on the MOOSE series. The following figure provides a structural overview of this research line, with arrows indicating major research connections such as conceptual influence, methodological borrowing, or task-enabling relationships.

The MOOSE Research Line

Like the moose that ventures into uncharted wilderness, the MOOSE series explores the untamed landscape of scientific hypotheses to uncover hidden insights.

Below are the main works in this line and their contributions.

  • MOOSE (ACL 2024)
    • The first work to show that LLMs can generate novel and valid scientific hypotheses.
  • MOOSE-Chem (ICLR 2025)
    • Provides a mathematically grounded theoretical foundation for automated scientific hypothesis discovery.
    • The first work to show that LLMs can rediscover the main innovations behind many research hypotheses published in Nature or Science.
  • MOOSE-Chem2 (NeurIPS 2025)
    • Introduces the task of fine-grained scientific hypothesis discovery, aiming to generate experimentally actionable hypotheses.
    • Frames the task as an optimization problem, and proposes hierarchical heuristic search to theoretically smooth the optimization landscape and reach better local optima.
  • MOOSE-Chem3
    • Introduces the task of experiment-guided ranking, bridging automated scientific hypothesis discovery and experimental feedback.
    • Proposes an experimental simulator that enables scalable research on experiment-guided ranking without relying on real wet-lab experiments.
  • MOOSE-Star
    • The first training recipe that enables tractable and scalable learning of P(hypothesis | research background).
    • Provides an inference recipe that supports more scalable test-time inference for discovery.
  • ResearchBench
    • The first large-scale benchmark for evaluating LLMs with a sufficient set of sub-tasks of scientific discovery: inspiration retrieval, hypothesis composition, and hypothesis ranking.
    • Suggests that LLMs can serve as research hypothesis mines, where stronger LLMs act as richer mines and greater inference compute enables more miners.
  • Survey
    • The first comprehensive survey of how LLMs can assist scientific research.

News

[2025.09]. MOOSE-Chem2 is accepted to NeurIPS 2025. Thanks to all my collaborators!

[2025.09]. I’ll host a tutorial on Agentic AI for Scientific Discovery at AAAI 2026 on January 21th, stay tuned!

[2025.05]. I’ll host a tutorial on Critical Works on LLMs for Scientific Discovery at AI4X on July 7th!

[2025.03]. I will take an invited talk at ICLR 2025 Agentic AI4S workshop on MOOSE-Chem on April 27th! The recording can be found in here.

[2025.02]. Got one paper accepted to CVPR 2025. Congrats to Di, Junxian, Jingdi, and Xunzhi!

[2025.01]. I’m invited as an Area Chair in ACL Rolling Review (ARR)!

[2025.01]. MOOSE-Chem is accepted to ICLR 2025. Thanks to all my collaborators!

[2024.07]. MOOSE has won the Best Poster Award in ICML 2024 AI4Science workshop!

[2024.05]. MOOSE is accepted to ACL 2024. Thanks to all my collaborators!

[2024.04]. I will take an invited talk at IJCAI 2024 AI4Research workshop on MOOSE on August 5th!

[2024.01]. Got one paper accepted to EACL 2024. Thanks to all my collaborators!

[2023.10]. Got one paper accepted to EMNLP 2023. Congrats to Wei!

[2023.06]. I will take an invited talk at ICCBR 2023 TMG workshop on our Case-based Reasoning paper on July 17th!

[2023.05]. Got one paper accepted to ACL 2023. Congrats to Jinjie!

[2023.04]. Our EACL 2023 paper will have an oral presentation!

[2023.01]. Got one paper accepted to EACL 2023. Thanks to all my collaborators!

[2020.10]. Got one paper accepted to EMNLP 2020 (findings). Thanks to all my collaborators!

Academic Services

Area Chair:

  • ARR

Conference Reviewer:

  • ARR, ICML 2026, NeurIPS 2025, COLM 2025, ICLR 2025, NLPCC 2024, COLM 2024, COLING 2024, EMNLP 2023, ACL 2023, EMNLP 2022, COLING 2022

Student Volunteer:

  • EMNLP 2023