About Me
I am Wanhao Liu (刘万昊), a Ph.D. student in Information and Communication Engineering at the 
University of Science and Technology of China, advised by Prof. Wanli Ouyang and Prof. Houqiang Li. I also conduct AI for Science research with the 
Shanghai AI Laboratory, where I am co-advised by Dr. Yuqiang Li. I am currently a research intern in the Foundation Model Team of Tencent TEG through the Qingyun Program.
Prior to this, I studied Polymer Physics and Chemistry at USTC, where I worked at the Hefei National Research Center for Physical Sciences at the Microscale with Prof. Guangming Liu, before transferring directly into the Ph.D. program. I received my B.S. in Chemistry (National Science Base Program) from 
Ocean University of China.
My research interests focus on LLMs for Scientific Discovery, Multimodal LLM Evaluation, and AI for Materials Science.
Collaboration: I am always happy to discuss AI for Science, automated scientific discovery, and LLM evaluation. If our interests overlap, please feel free to contact me.
🔥 News
🎉 Our work on LLMs as designers for anisotropic polymer thermocells (OmniMat) was published in Cell Reports Physical Science and selected as a cover candidate.
💼 Joined the Foundation Model Team of Tencent TEG as a Qingyun Program research intern, working on LLM evaluation and self-evolving evaluation systems.
🚀 Released OmniMatBench, a human-calibrated multimodal benchmark spanning 19 materials-science subfields, now integrated into OpenCompass.
🎉 Our chemical reasoning model Chem-R was accepted to KDD 2026.
📣 PolyReal, our benchmark for real-world polymer science workflows (CVPR 2026 Findings), is now available on arXiv and Hugging Face.
🎉 MOOSE-Chem2 was accepted to NeurIPS 2025, and MOOSE-Chem3 to the NeurIPS 2025 AI for Science Workshop.
🎉 MOOSE-Chem, showing that LLMs can rediscover unseen chemistry hypotheses, was accepted to ICLR 2025.
📣 Joined the AI for Science group at the Shanghai AI Laboratory as a research intern.
📝 Selected Publications (Full List)
* Equal contribution. † Corresponding author.
LLMs for Scientific Discovery
MOOSE-Chem: Large Language Models for Rediscovering Unseen Chemistry Scientific Hypotheses
- The first work showing that LLMs can rediscover hypotheses published in Nature or Science without data contamination, by retrieving non-obvious, cross-domain inspirations and iteratively evolving candidate hypotheses.
MOOSE-Chem2: Exploring LLM Limits in Fine-Grained Scientific Hypothesis Discovery via Hierarchical Search
- A chemistry-informed five-level hierarchy and hierarchical search workflow that refine coarse hypotheses into experimentally actionable designs.
MOOSE-Chem3: Toward Experiment-Guided Hypothesis Ranking via Simulated Experimental Feedback
- Introduces experiment-guided hypothesis ranking with a simulator validated on 124 chemistry hypotheses with reported outcomes; an LLM policy that clusters hypotheses by functional role and learns from simulated feedback outperforms pre-experiment baselines. Presented at the NeurIPS 2025 AI for Science Workshop.
Large Language Models as Designers for Autonomous Research of Anisotropic Polymer Thermocells
- OmniMat, an LLM-driven framework for experimental protocol design. The AI-designed anisotropic polymer thermocell, validated in the wet lab, reaches 0.68 mW m−2 K−2 normalized power density with toughness above 21 MJ m−3. Selected as a cover candidate.
LLM Training and Reasoning Optimization
Multimodal LLM Evaluation and Benchmarking
PolyReal: A Benchmark for Real-World Polymer Science Workflows
- The first multimodal benchmark covering the full polymer research workflow, from knowledge and lab safety to mechanism reasoning, raw-data analysis, and application exploration; 15 frontier MLLMs reveal a large gap between knowing and applying science.
OmniMatBench: A Human-Calibrated Multimodal Reasoning Benchmark Across 19 Materials Science Subfields
- 3,171 expert-calibrated QA and calculation problems across 19 materials-science subfields, integrated into OpenCompass; 13 frontier MLLMs show persistent gaps in materials reasoning and scientific execution.
Other Publications
- Developing ChemMLLM as a Multimodal Large Language Model for Chemistry
Cell Reports Physical Science, 2026 [arXiv] [BIB] - LoRA-Chem: Modular Machine Learning for Multitask Prediction in Organic Reactions
CCS Chemistry, 2026 [BIB] - SkillsInjector: Dynamic Skill Context Construction for LLM Agents
arXiv, 2026 - MolAct: An Agentic RL Framework for Molecular Editing and Property Optimization
arXiv, 2025 - Probing Scientific General Intelligence of LLMs with Scientist-Aligned Workflows
arXiv, 2025
🏅 Honors and Awards
- 2025-2026 Outstanding Academic Report Nomination Award, USTC Doctoral Student Academic Forum
- 2023-2024 Outstanding Communist Youth League Member, USTC
- 2021-2022 First-Class Scholarship, Hefei National Research Center for Physical Sciences at the Microscale
- 2021-2022 First-Class Academic Scholarship, USTC
- 2020-2021 Outstanding Graduate and Outstanding Student, Ocean University of China
📖 Educations
- 2023.09 - 2027.06 (expected), Ph.D. in Information and Communication Engineering, University of Science and Technology of China, joint program with Shanghai AI Laboratory
- Supervised by Prof. Wanli Ouyang, Prof. Houqiang Li, and Dr. Yuqiang Li
- 2021.09 - 2023.09, Graduate study in Polymer Physics and Chemistry (transferred to the Ph.D. program), University of Science and Technology of China, Hefei, China
- Hefei National Research Center for Physical Sciences at the Microscale, supervised by Prof. Guangming Liu
- 2017.09 - 2021.06, B.S. in Chemistry (National Science Base Program), Ocean University of China, Qingdao, China
💻 Internships
- 2026.09 - Present, Qingyun Program Research Intern, Foundation Model Team, Technology and Engineering Group (TEG), Tencent [link]
- Large language model evaluation and self-evolving evaluation systems
- 2024.07 - 2026.09, AI4Science Research Intern, Shanghai AI Laboratory [link]
- LLMs for automated scientific discovery; benchmark construction for chemistry and materials science
🎤 Talks
- 2026.09, Lingxiao Club Tech Talk, Shanghai AI Laboratory
- 2025.06, The 6th “Dechuang · Cutting-Edge Innovation” Doctoral Student Academic Forum, USTC
📋 Academic Services
- Conference Reviewer: ACL Rolling Review (ARR), NeurIPS, ECCV