|
Zhongzheng Li I am a second-year PhD student at CASIA (Institute of Automation, Chinese Academy of Sciences), advised by Prof. Xiaoguang Zhao. I received my BS from Zhejiang University. My research interests mainly revolve around LLM and RL. I am always open to research discussions and collaborations. Email / Google Scholar / Github / CV |
|
PreprintsPapers marked with * indicate equal contribution. Highlighted papers are first-authored or co-first-authored. |
|
OPTScientist: Multi-Agent Discovery of Typed Optimizer Programs for Transformer Pretraining
Zhongzheng Li, Tiancan Feng, Wenhao Li, Qingsong Ran, Shikun Feng, Xiaoyuan Zhang, Yue Wang, Xiaoguang Zhao (Conference Submission) A theory-guided multi-agent framework that frames optimizer design as a scientific search over a typed DSL, with Theorist, Designer, Engineer, and Reviewer agents collaborating in a closed loop. It discovers RS-MR, a reduced-state matrix optimizer that improves transformer pretraining over strong baselines. |
|
WMLLM: Self-Evolving Optimization Agents via Predict-Then-Act World Modeling
Zhongzheng Li, Qingsong Ran, Shikun Feng, Nian Ran, Wenhao Li, Xiaoyuan Zhang, Yue Wang, Xiaoguang Zhao (Conference Submission) A self-evolving optimization agent built on predict-then-act world modeling: it first predicts promising directions, then acts to generate candidates. Combining multi-turn refinement, population-based search, and reinforcement learning, WMLLM achieves state-of-the-art sample efficiency on multi-objective molecular optimization. |
|
LingDE: Linguistic Operators for Differential Evolution with Large Language Models
Qingsong Ran, Zhongzheng Li, Shikun Feng, Xiaoyuan Zhang, Jinbiao Nie, Nian Ran, Wenhao Li, Yue Wang, Wanjing Ma (Conference Submission) Linguistic operators for differential evolution that leverage LLMs to perform semantic mutation and crossover on natural language instructions, enabling effective optimization in combinatorial and black-box problem settings. |
|
MCCE: A Framework for Multi-LLM Collaborative Co-Evolution
Nian Ran*, Zhongzheng Li*, Yue Wang, Qingsong Ran, Xiaoyuan Zhang, Shikun Feng, Richard Allmendinger, Xiaoguang Zhao ICML, 2026 arXiv / code A framework for multi-LLM collaborative co-evolution that leverages multi-objective optimization and DPO training to enable multiple LLMs to collaboratively improve through experience sharing and synthesis. |
|
ExLLM: Experience-Enhanced LLM Optimization for Molecular Design and Beyond
Nian Ran, Yue Wang, Xiaoyuan Zhang, Zhongzheng Li, Qingsong Ran, Wenhao Li, Richard Allmendinger arXiv, 2025 arXiv / code An experience-enhanced LLM optimization framework for molecular design that accumulates and leverages past optimization experiences to guide future generation, achieving superior performance on molecular design benchmarks. |
Experience |
| Sep. 2024 – Present |
CASIA, Institute of Automation, Chinese Academy of Sciences PhD Student, supervised by Prof. Xiaoguang Zhao |
| Sep. 2020 – Jun. 2024 |
Zhejiang University Undergraduate Student, GPA: 3.92/4.0 |
Awards |
|
|
Template from Jon Barron. |