Suozhi Huang ☕️
Suozhi Huang

Grad Student Researcher

Yao Class, Tsinghua University

About Me

Hi, I’m Suozhi, first year Ph.D. student in Princeton University studying in ECE department. My current interest is in LLM agents which can solve specific hard tasks that human can’t even handle.

I did my undergrad Computer Science(major in AI) at Yao Class in Tsinghua University. My past research interests included automated theorem proving, language model reasoning and AI for science(including math, physics and biochem). I am fortunate to work with Prof. Anima Anandkumar at Caltech. I am also a member of Internlm-Math team, developing llm formal math provers. I start my research journey from robotics.

Download CV
Interests
  • Artificial Intelligence
  • Automated Theorem Proving
  • Reasoning in LLMs
Education

  • B.E. Computer Science (Major in AI)

    Yao Class, Tsinghua University

📚 My Research

My research interests include LLMs and agentic tasks. Additionally, I aim to incorporate reinforcement learning (RL) and real world high dimensional data to optimize the agent strategy, enabling systems to learn from expert strategies and improve their performance over time.

My goal is to create robust frameworks that leverage LLM capabilities for hard decision problems.

Please reach out to collaborate 😃

Publications

View all publications on Google Scholar ↗

2026

Any2Poster: Any-Source Poster Generation Across Modalities and Domains
Amogh Vinaykumar, Aiden Li, Suozhi Huang, and Shilong Liu
arXiv preprint · Paper · Code
FutureX-Pro: Extending Future Prediction to High-Value Vertical Domains
Jiashuo Liu, Siyuan Chen, Zaiyuan Wang, Zhiyuan Zeng, Jiacheng Guo, Liang Hu, Lingyue Yin, Suozhi Huang, et al.
arXiv preprint · Paper

2025

CryptoBench: A Dynamic Benchmark for Expert-Level Evaluation of LLM Agents in Cryptocurrency
Jiacheng Guo, Suozhi Huang, Zixin Yao, Yifan Zhang, Yifu Lu, Jiashuo Liu, et al.
arXiv preprint · Paper
Gated Attention for Large Language Models: Non-linearity, Sparsity, and Attention-Sink-Free
Zihan Qiu, Zekun Wang, Bo Zheng, Zeyu Huang, Kaiyue Wen, Songlin Yang, Rui Men, Le Yu, Fei Huang, Suozhi Huang, Dayiheng Liu, Jingren Zhou, and Junyang Lin
NeurIPS 2025 · Paper
LeanProgress: Guiding Search for Neural Theorem Proving via Proof Progress Prediction
Suozhi Huang, Peiyang Song, Robert Joseph George, and Anima Anandkumar
Transactions on Machine Learning Research · Paper

2024

InternLM2.5-StepProver: Advancing Automated Theorem Proving via Expert Iteration on Large-Scale LEAN Problems
Zijian Wu*, Suozhi Huang*, Zhejian Zhou, Huaiyuan Ying, Jiayu Wang, Dahua Lin, and Kai Chen
Technical report · Paper · Code · Dataset
A Knowledge-Data Dual-Driven Framework for Predicting the Molecular Properties of Rechargeable Battery Electrolytes
Yuchen Gao, Yu-Hang Yuan, Suozhi Huang, Nan Yao, Legeng Yu, Yao-Peng Chen, Qiang Zhang, and Xiang Chen
Angewandte Chemie International Edition · Paper
ActFormer: Scalable Collaborative Perception via Active Queries
Suozhi Huang*, Juexiao Zhang*, Yiming Li, and Chen Feng
ICRA 2024 · Paper · Code · Project

Experience

  1. Researcher in Caltech Anima AI+Science lab

    Caltech

    Responsibilities include:

    • Hold a new research project
    • Create and integrate plugin for LeanCopilot

  2. Researcher in Internlm-Math team

    Shanghai AI lab (PJLab)

    Responsibilities include:

    • Built critic model for expert iteration on Lean dataset(Lean workbook & Lean github)
    • Replaced Best first search method with critic model guided search in tactic selection
    • Contruct a new comprehensive dataset by expert iteration searching
    • Finished one SOTA paper in LLM for theorem proving

  3. Researcher in NYU AI4CE Lab

    NYU AI4CE Lab

    Responsibilities include:

    • Proposed a scalable, active paradigm in query-based camera-only collaboration task, which reduced the cost in feature transfer
    • Implemented a pose-guided selection network in deformable attention in camera-based perception, which greatly reduced the number of BEV queries with effective selection
    • Completed all experiments in multi-agent dataset, analyzed the results, and improved scalability
    • Submitted a paper to ICRA 2024 as first author

Education

  1. B.E. Computer Science (Major in AI)

    Yao Class, Tsinghua University

    GPA: 3.6/4.0

    Courses included:

    • machine learning
    • quantum computing
    • autonomous driving