Zexiong Ma

Zexiong Ma

C.S., Peking University

Zexiong Ma is a fourth-year Ph.D. candidate supervised by Prof. Bing Xie in Software Engineering Institute, School of Computer Science, Peking University. Previously, He received B.S. degree in Computer Science from Tianjin University in 2021.

His research interest is the application of large language models in coding and reasoning tasks, including:

  • (1) coding: library-oriented code generation, retrievel-augmented code generation, and coding agent;
  • (2) reasoning: mathematical reasoning, long-context reasoning and tool-interative reasoning.

Recently, he has been fully devoted to rule-based reinforcement learning to enhance the reasoning and coding capabilities of LLMs. Please feel free to email him if you’re interested in his research or simply want to talk!

📝 Publications

  • [ICSE 2026] Evaluating Generated Commit Messages with Large Language Models.
    Qunhong Zeng, Yuxia Zhang, Zexiong Ma, Bo Jiang, Ningyuan Sun, Klaas-Jan Stol, Xingyu Mou, Hui Liu.
  • [ACL 2025] SoRFT: Issue Resolving with Subtask-oriented Reinforced Fine-Tuning. [Paper]
    Zexiong Ma, Chao Peng, Pengfei Gao, Xiangxin Meng, Yanzhen Zou, Bing Xie.
  • [LCFM@ICML 2025] Enhancing Retrieval-Augmented Generation with Dehallucinating Parallel Context Extension. [Paper]
    Zexiong Ma, Shengnan An, Zeqi Lin, Yanzhen Zou, Jian-Guang Lou, Bing Xie.
  • [NeurIPS 2024] Make Your LLM Fully Utilize the Context. [Paper] [Code]
    Shengnan An, Zexiong Ma, Zeqi Lin, Nanning Zheng, Jian-Guang Lou, Weizhu Chen.
  • [EMNLP 2024] Can LLMs Learn From Mistakes? An Empirical Study on Reasoning Tasks. [Paper] [Code]
    Preprint Version: Learning From Mistakes Makes LLM Better Reasoner. [Paper]
    Shengnan An, Zexiong Ma, Siqi Cai, Zeqi Lin, Nanning Zheng, Jian-Guang Lou, Weizhu Chen.
  • [ICPC 2024] Compositional API Recommendation for Library-Oriented Code Generation. [Paper]
    Zexiong Ma, Shengnan An, Bing Xie, Zeqi Lin.
  • [preprint] Repository Structure-Aware Training Makes SLMs Better Issue Resolver. [Paper]
    Zexiong Ma, Shengnan An, Zeqi Lin, Yanzhen Zou, Bing Xie.
  • [preprint] An Empirical Study on LLM-based Agents for Automated Bug Fixing. [Paper]
    Xiangxin Meng, Zexiong Ma, Pengfei Gao, Chao Peng

📖 Educations

  • 2021.09 - 2026.07 (expected), Peking University, Ph.D. in Computer Science. Adviser: Prof. Bing Xie.
  • 2017.09 - 2021.07, Tianjin University, B.S. Degree in Computer Science.

🎖 Honors and Awards

  • 2019, National Scholarship.
  • 2019, National Student Computer System Design Capability Challenge (Loongson Cup), Third Prize.
  • 2018, International Collegiate Programming Contest (ACM/ICPC), Silver Model.
  • 2015, National Olympiad in Informatics in Provinces (NOIP), First Prize.

📌 Academic Services

Reviewer for ICLR, LCFM@ICML, ACL Rolling Review

💻 Internships

  • 2024.10 - now, Trae/Marscode Research, ByteDance, China. Work with Chao Peng, Pengfei Gao, and Xiangxin Meng.
  • 2023.03 - 2024.09, DKI Group, Microsoft, China. Work with Zeqi Lin, and Shengnan An.