Zhenghai Xue

Tencent Hy. zhenghai001@e.ntu.edu.sg

prof_pic.jpg

I am a member of technical staff at Tencent Hy, in charge of production-level post training with OPD and RL. I received my Ph.D. from the College of Computing and Data Science, Nanyang Technological University, Singapore, supervised by Prof. Bo An. Previously, I obtained my B.Sc. of Artificial Intelligence from Nanjing University in 2022. In my undergraduate study, I worked with Prof. Yang Yu at LAMDA and Prof. Bolei Zhou at MMLab.

I have been fortunate to work as a research intern at the following companies:

  • Moonshot AI with Dr. Chenjun Xiao. We worked on improving the long-horizon instruction-following abilities of coding agents.
  • TikTok with Dr. Qian Liu. We worked on scalable and stable algorithms for agentic RL training.
  • Kuaishou Technology with Dr. Qingpeng Cai. We worked on RL for improving user retention.
  • Kunlun 2050 Research with Prof. Shuicheng Yan. We worked on robust and human-in-the-loop RL.

News

Jul 4, 2025 We release SimpleTIR, an end-to-end solution for stable multi-turn tool use RL training.
May 3, 2025 One paper is accepted at ICML 2025 as Spotlight Poster!
Jan 23, 2025 One paper is accepted at WWW 2025. Two papers are accepted to ICLR 2025.
Nov 8, 2023 I will give a talk on Optimizing Long-term User Engagement in the Applied Artificial Intelligence Workshop of DAI 2023!
Oct 19, 2023 Our paper “State Regularized Policy Optimization on Data with Dynamics Shift” is accepted by NeurIPS 2023!

Selected publications

  1. SimpleTIR: End-to-End Reinforcement Learning for Multi-Turn Tool-Integrated Reasoning
    Zhenghai Xue*, Longtao Zheng*, Qian Liu, Yingru Li, Xiaosen Zheng, Zejun MA, and Bo An
    In The Thirteenth International Conference on Learning Representations, 2026
    arXiv HTML Code
  2. Group-in-Group Policy Optimization for LLM Agent Training
    Lang Feng, Zhenghai Xue, Tingcong Liu, and Bo An
    Advances in Neural Information Processing Systems, 2025
    arXiv Code
  3. Policy Optimization under Imperfect Human Interactions with Agent-Gated Shared Autonomy
    Zhenghai Xue, Bo An, and Shuicheng Yan
    In The Thirteenth International Conference on Learning Representations, 2025
    HTML
  4. Policy Regularization on Globally Accessible States in Cross-Dynamics Reinforcement Learning
    Zhenghai Xue, Lang Feng, Jiacheng Xu, Kang Kang, Xiang Wen, Bo An, and Shuicheng Yan
    In Forty-Second International Conference on Machine Learning (Spotlight), 2025
    arXiv
  5. AURO: Reinforcement Learning for Adaptive User Retention Optimization in Recommender Systems
    Zhenghai Xue, Qingpeng Cai, Tianyou Zuo, Bin Yang, Lantao Hu, Peng Jiang, Kun Gai, and Bo An
    In Proceedings of the ACM Web Conference (Oral), 2025
    arXiv
  6. State Regularized Policy Optimization on Data with Dynamics Shift
    Zhenghai Xue, Qingpeng Cai, Shuchang Liu, Dong Zheng, Peng Jiang, Kun Gai, and Bo An
    Advances in Neural Information Processing Systems, 2023
    arXiv HTML Code
  7. Guarded Policy Optimization with Imperfect Online Demonstrations
    Zhenghai Xue, Zhenghao Peng, Quanyi Li, Zhihan Liu, and Bolei Zhou
    In The Eleventh International Conference on Learning Representations (Spotlight), 2023
    arXiv HTML Code
  8. Regret minimization experience replay in off-policy reinforcement learning
    Xu-Hui Liu*, Zhenghai Xue*, Jingcheng Pang, Shengyi Jiang, Feng Xu, and Yang Yu
    Advances in Neural Information Processing Systems, 2021
    arXiv Code Video