English

邓贺烜

Hexuan Deng 邓贺烜

photo

GitHub Google Scholar
ORCID Semantic Scholar

我是哈尔滨工业大学（深圳）计算机科学与技术专业博士生，并在北京中关村学院联培。 研究兴趣主要包括大语言模型、数据中心人工智能、强化学习和智能体。

教育经历

研究经历

研究工作

* 表示共同一作。

会议论文

  1. CoCoReviewBench: A Completeness- and Correctness-Oriented Benchmark for AI Reviewers.
    Hexuan Deng*, Xiaopeng Ke*, Yichen Li*, Ruina Hu*, Dehao Huang, Derek F. Wong, Yue Wang, Xuebo Liu, Min Zhang.
    ICML, 2026. [arXiv] [Github]
  2. REA-RL: Reflection-Aware Online Reinforcement Learning for Efficient Large Reasoning Models.
    Hexuan Deng, Wenxiang Jiao, Xuebo Liu, Jun Rao, Min Zhang.
    ICLR, 2026. [arXiv] [Github]
  3. DRPruning: Efficient Large Language Model Pruning through Distributionally Robust Optimization.
    Hexuan Deng, Wenxiang Jiao, Xuebo Liu, Min Zhang, Zhaopeng Tu.
    ACL Main Conference, 2025. [arXiv] [Github]
  4. NewTerm: Benchmarking Real-Time New Terms for Large Language Models with Annual Updates.
    Hexuan Deng, Wenxiang Jiao, Xuebo Liu, Min Zhang, Zhaopeng Tu.
    NeurIPS D&B, 2024. [arXiv] [Github]
  5. Improving Simultaneous Machine Translation with Monolingual Data.
    Hexuan Deng, Liang Ding, Xuebo Liu, Meishan Zhang, Dacheng Tao, Min Zhang.
    AAAI, 2023. [arXiv] [Github]
  6. MemoNoveltyAgent: A Historical-Memory-Augmented Multi-Agent System for Paper Novelty Assessment.
    JiaJun Hou*, Hexuan Deng*, Wenxiang Jiao, Xuebo Liu, Xiaopeng Ke, Derek F. Wong, Min Zhang.
    EMNLP Findings, 2026. [Github]
  7. RouterKGQA: Specialized–General Model Routing for Constraint-Aware Knowledge Graph Question Answering.
    Bo Yuan*, Hexuan Deng*, Xuebo Liu, Long Bang, Zheng Shuai, Zhou Mingzhen, Min Zhang.
    AACL-IJCNLP Main Conference, 2026. [Github]
  8. Dynamic Sampling that Adapts: Self-Aware Iterative Data Persistent Optimization for Mathematical Reasoning.
    Jun Rao, Xuebo Liu, Hexuan Deng, Zepeng Lin, Zixiong Yu, Jiansheng Wei, Xiaojun Meng, Min Zhang.
    Findings of ACL, 2026. [arXiv]
  9. AQuilt: Weaving Logic and Self-Inspection into Low-Cost, High-Relevance Data Synthesis for Specialist LLMs.
    Xiaopeng Ke, Hexuan Deng, Xuebo Liu, Jun Rao, Zhenxi Song, Jun Yu, Min Zhang.
    EMNLP Main Conference, 2025. [arXiv] [Github]

预印本

  1. GRPODropout: Less is More for Online Reinforcement Learning Rollouts.
    Hexuan Deng*, Zihao Yan*, Xuebo Liu, Shuo Nie, Yue Wang, Chen Wang, Zhaohua Zhang, Tianwen Jiang, Qiuyong Xiao, Jihong Zhang, Min Zhang.
    arXiv preprint, 2026. [Github]
  2. SWE-Journey: Towards More Realistic Evaluation of Coding Assistants through Long-Horizon, Multi-Turn Interaction.
    Hexuan Deng, Yue Wang, Wenyu Jiang, Cheng Yang, Haolin Yang, Zhaohua Zhang, Chenchen Zhao, Beiduo Chen, Muxi Chen, Sa Zhu, Geyuan Zhu, Jianhuan Zhuo, Qiuyong Xiao, Tianwen Jiang, Jihong Zhang, Xuebo Liu.
    arXiv preprint, 2026.
  3. Stop Rewarding Hallucinated Steps: Faithfulness-Aware Step-Level Reinforcement Learning for Small Reasoning Models.
    Shuo Nie*, Hexuan Deng*, Chao Wang, Ruiyu Fang, Xuebo Liu, Shuangyong Song, Yu Li, Min Zhang, Xuelong Li.
    arXiv preprint, 2026. [Github]
  4. Implicit Compression Regularization: Concise Reasoning via Internal Shorter Distributions in RL Post-Training.
    Chen Wang, Hexuan Deng, Yining Zhang, Yuchen Zhang, Jionghao Bai, Zhaochun Li, Ge Lan, Yue Wang.
    arXiv preprint, 2026.

项目经历

  1. Co-supervision
    对齐与深度推理小组负责人, Since May 2025
    指导本科生和硕士生开展数据合成、强化学习与多智能体协作研究，覆盖idea、实验设计与论文写作，独立组织组会并参与高水平论文产出。
  2. 招聘大模型
    AI辅助面试负责人, August 2024 -- April 2025
    带领六人团队，专注于个性化定制笔试和面试题、结果评价与岗位匹配，并通过类长文本Agent和辩论Agent自动更新题库。
  3. Holistic Exploration on Universal Decompositional Semantic Parsing: Architecture, Data Augmentation, and LLM Paradigm.
    Hexuan Deng, Xin Zhang, Meishan Zhang, Xuebo Liu, Min Zhang.
    ACL Workshop, 2024. [arXiv] [Github]
    Focused on improving model performance and inference speed for UDS semantic parsing tasks.
  4. "LiZhi" Foundation Model and Intelligent Dialogue System
    Team Member, April 2023 -- June 2023
    Focused on enhancing the model's mathematical capabilities. Introduced a calculator tool to significantly reduce calculation errors.

程序委员会 / 审稿人