Qingyao Li

About

I am a Ph.D. candidate in Computer Science and Technology at Shanghai Jiao Tong University, advised by Prof. Weinan Zhang and Prof. Yong Yu. I am also a member of the Zhiyuan Honorary Doctoral Program.

My research focuses on large language model reasoning and code intelligence. I develop search- and learning-based algorithms, including Monte Carlo Tree Search and reinforcement learning, that integrate structured test-time reasoning with fine-grained execution feedback. The goal is to make LLM-based code agents more robust on complex programming tasks.

Previously, I worked on intelligent education, especially reinforcement learning for learning path recommendation. That experience continues to shape how I think about sequential decision making, optimization, and adaptive feedback.

News

  • Apr 2026 AdverMCTS is available as an ICML 2026 paper and arXiv preprint.
  • Mar 2026 Started a research internship on agentic reinforcement learning and general agents.
  • Sep 2025 Began a Ph.D. exchange at Nanyang Technological University with Prof. Bo An.
  • May 2025 Received the Huawei Excellent Intern award.

Research Interests

  • LLM reasoning for code: test-time search, process verification, debugging, and execution feedback.
  • Search and learning: MCTS, adversarial training, reinforcement learning, and model improvement from verification signals.
  • Sequential decision making: reinforcement learning for adaptive recommendation and educational decision making.

Publications

* denotes equal contribution. Author names linked to my work are in bold.

2026

MCTS tree with one branch flagged as adversarial — AdverMCTS

AdverMCTS: Combating Pseudo-Correctness in Code Generation via Adversarial Monte Carlo Tree Search

Qingyao Li, Weiwen Liu, Weinan Zhang, Yong Yu, Bo An

ICML 2026

Adversarial MCTS that couples code search with active vulnerability discovery to reduce pseudo-correctness.

paper arXiv

Solver and adversary in a min-max loop — ATGen

ATGen: Adversarial Reinforcement Learning for Test Case Generation

Qingyao Li, Xinyi Dai, Weiwen Liu, Xiangyang Li, Yasheng Wang, Ruiming Tang, Yong Yu, Weinan Zhang

ICLR 2026

An adversarial RL framework for generating challenging tests and stronger reward signals for code models.

paper arXiv

Knowledge graph with central concept node — Concept Recommendation

Learning Structure and Knowledge Aware Representation with Large Language Models for Concept Recommendation

Qingyao Li, Wei Xia, Kounianhua Du, Qiji Zhang, Weinan Zhang, Ruiming Tang, Yong Yu

AAMAS 2026 Extended Abstract

LLM-based concept representations with a graph-based adapter for knowledge-aware recommendation.

arXiv

2025

MCTS tree with rethink loop on an erroneous thought — RethinkMCTS

RethinkMCTS: Refining Erroneous Thoughts in Monte Carlo Tree Search for Code Generation

Qingyao Li, Wei Xia, Kounianhua Du, Xinyi Dai, Ruiming Tang, Yasheng Wang, Yong Yu, Weinan Zhang

EMNLP 2025

Thought-level MCTS with fine-grained execution feedback for correcting reasoning paths before code generation.

paper arXiv

Code block with a natural language repair sketch — NL-Debugging

NL-Debugging: Exploiting Natural Language as an Intermediate Representation for Code Debugging

Weiming Zhang*, Qingyao Li*, Xinyi Dai, Jizheng Chen, Kounianhua Du, Weiwen Liu, Yasheng Wang, Ruiming Tang, Yong Yu, Weinan Zhang

EMNLP 2025

Natural language sketches as an intermediate representation for locating and repairing algorithmic flaws.

paper arXiv

Chain of thought steps with process reward scores — CodePRM

CodePRM: Execution Feedback-enhanced Process Reward Model for Code Generation

Qingyao Li, Xinyi Dai, Xiangyang Li, Weinan Zhang, Yasheng Wang, Ruiming Tang, Yong Yu

ACL 2025 Findings

A process reward model that uses execution feedback to score thought steps and guide generate-verify-refine search.

paper ACL Anthology

2024 and Earlier

Privileged teacher distilling knowledge to an RL student policy — PFD

Privileged Knowledge State Distillation for Reinforcement Learning-based Educational Path Recommendation

Qingyao Li, Wei Xia, Liang Yin, Jiarui Jin, Yong Yu

KDD 2024

Privileged feature distillation for stabilizing reinforcement learning in learning path recommendation.

paper ACM DL

High-level policy steering a graph of low-level concepts — GEHRL

Graph Enhanced Hierarchical Reinforcement Learning for Goal-oriented Learning Path Recommendation

Qingyao Li, Wei Xia, Li'ang Yin, Jian Shen, Renting Rui, Weinan Zhang, Xianyu Chen, Ruiming Tang, Yong Yu

CIKM 2023

Graph-enhanced hierarchical RL for decomposing learning goals and recommending concept-level learning paths.

paper ACM DL

Open book topped with a small neural-network glyph — LLM for Education

Adapting Large Language Models for Education: Foundational Capabilities, Potentials, and Challenges

Qingyao Li, Lingyue Fu, Weiming Zhang, Xianyu Chen, Jingwei Yu, Wei Xia, Weinan Zhang, Ruiming Tang, Yong Yu

Preprint

A survey of LLM capabilities and adaptation methods for education.

paper arXiv

Experience & Education

2026.3 – present

Research Intern, Xiaohongshu

Agentic reinforcement learning and general agent research.

2025.9 – 2026.3

Ph.D. Exchange, Nanyang Technological University

College of Computing and Data Science. Advisor: Prof. Bo An.

2024.7 – 2025.5

Student Researcher, Huawei Noah's Ark Lab

Research on LLM code generation and reinforcement learning. Mentors: Wei Xia, Xinyi Dai, Ruiming Tang.

2022.9 – present

Ph.D. Candidate, Shanghai Jiao Tong University

Computer Science and Technology, Zhiyuan Honorary Doctoral Program. Advisors: Prof. Weinan Zhang and Prof. Yong Yu.

2018.9 – 2022.8

B.S., Xi'an Jiao Tong University

Automation, Qian Xuesen Class. GPA: 4.05/4.3, ranked 1/25.

Honors, Service & Patents

  • Huawei Excellent Intern, 2025
  • Outstanding Graduate of Xi'an Jiao Tong University, 2022
  • International Excellence Award in Baidu Big Data Competition, 2020
  • Conference reviewer: ICML 2026, KDD 2024, CIKM 2023, AAAI 2023
  • Concept recommendation system integrating human knowledge structures, 2024
  • A wireless charger for pacemakers, utility model patent, 2022

Contact

I am interested in collaborations on LLM reasoning, code agents, reinforcement learning, and robust evaluation for programming tasks.

ly890306@sjtu.edu.cn CV