万海 (教授)

研究方向

• 神经符号学习 (Neuro-Symbolic Learning)

  

  该方向研究如何用可微的神经网络学习和推理离散的符号对象，包括有限迹/无限迹线性时序逻辑（LTL/LTLf）公式、正则表达式、有限状态自动机与有序二元决策图等。核心思路是设计能够忠实模拟符号推理过程的神经编码，把离散的结构搜索转化为连续的参数优化，从而在含噪声数据上稳健地学习出可解释的符号规约，或在多项式时间内近似求解可满足性检查、轨迹生成等原本计算困难的问题。近年还尝试将大语言模型的推理能力与逻辑推理迭代结合。这是当前的主要研究重心，成果发表于 AAAI、IJCAI、ISSTA、ASE 等会议。

  研究论文：
  • Weilin Luo, Tingchen Han, Junming Qiu, Hai Wan（万海）, Jianfeng Du, Bo Peng, Guohui Xiao, Yanan Liu: NADA: Neural Acceptance-Driven Approximate Specification Mining. ISSTA 2025: ISSTA079. [PDF] [Code]
  • Junming Qiu, Rongzhen Ye, Weilin Luo, Kunxun Qi, Hai Wan（万海）, Yue Yu: OBDD-NET: End-to-End Learning of Ordered Binary Decision Diagrams. CIKM 2025: 2430-2439. [PDF] [Code]
  • Hai Wan（万海）, Pingjia Liang, Jianfeng Du, Weilin Luo, Rongzhen Ye, Bo Peng: End-to-End Learning of LTLf Formulae by Faithful LTLf Encoding. AAAI 2024: 9071-9079. [PDF]
  • Weilin Luo, Pingjia Liang, Junming Qiu, Polong Chen, Hai Wan（万海）, Jianfeng Du, Weiyuan Fang: Learning to Check LTL Satisfiability and to Generate Traces via Differentiable Trace Checking. ISSTA 2024: 996-1008. [PDF] [Code]
  • Rongzhen Ye, Tianqu Zhuang, Hai Wan（万海）, Jianfeng Du, Weilin Luo, Pingjia Liang: A Noise-tolerant Differentiable Learning Approach for Single Occurrence Regular Expression with Interleaving. AAAI 2023: 4809-4817. [PDF]



---

• 知识图谱补全与关系抽取 (Knowledge Graph Completion and Relation Extraction)

  

  该方向研究知识图谱补全与（文档级）关系抽取，重点是逻辑规则的自动学习及其与神经网络的结合。工作包括在链式规则中引入实体类型约束和双向一致性以提升链接预测的准确性与可解释性，融合结构化三元组与文本知识进行归纳式补全，以及端到端地联合训练关系抽取模型与规则推理模块以缓解错误传播。相关成果发表于 NeurIPS、ACL、CIKM 等会议。

  研究论文：
  • Kunxun Qi, Jianfeng Du, Hai Wan（万海）: Bi-directional Learning of Logical Rules with Type Constraints for Knowledge Graph Completion. CIKM 2024: 1899-1908. [PDF] [Code]
  • Kunxun Qi, Jianfeng Du, Hai Wan（万海）: End-to-end Learning of Logical Rules for Enhancing Document-level Relation Extraction. ACL (1) 2024: 7247-7263. [PDF]
  • Kunxun Qi, Jianfeng Du, Hai Wan（万海）: Learning from Both Structural and Textual Knowledge for Inductive Knowledge Graph Completion. NeurIPS 2023. [PDF] [Code]
  • Huaping Zhong, Jianwen Zhang, Zhen Wang, Hai Wan, Zheng Chen: Aligning Knowledge and Text Embeddings by Entity Descriptions. EMNLP 2015: 267-272.



---

• 自然语言处理 (Natural Language Processing)

  

  该方向围绕文本理解与推理任务展开，涉及事实核查与证据检索、跨语言自然语言推理与基于方面的情感分析、隐喻检测，以及（多模态）社会关系抽取等。方法上既使用预训练语言模型、提示学习、小样本学习与量子神经网络，也引入深度强化学习和最小碰撞集等符号技术来挖掘声明的精确或全部最小证据。相关工作发表于 ACL、AAAI 及 Information Sciences 等。

  研究论文：
  • Yanan Liu, Hai Wan（万海）, Jianfeng Du, Yao Wang, Kunxun Qi, Weilin Luo: Learning to mine all minimal evidences for unverified claims. Information Sciences 720: 122505 (2025). [PDF]
  • Xingqiang Zhao, Hai Wan（万海）, Kunxun Qi: QPEN: Quantum Projection and Quantum Entanglement Enhanced Network for Cross-Lingual Aspect-Based Sentiment Analysis. AAAI 2024: 19670-19677. [PDF]
  • Kunxun Qi, Hai Wan（万海）, Jianfeng Du, Haolan Chen: Enhancing Cross-lingual Natural Language Inference by Prompt-learning from Cross-lingual Templates. ACL (1) 2022: 1910-1923. [PDF] [Code]
  • Hai Wan（万海）, Manrong Zhang, Jianfeng Du, Ziling Huang, Yufei Yang, Jeff Z. Pan: FL-MSRE: A Few-Shot Learning based Approach to Multimodal Social Relation Extraction. AAAI 2021: 13916-13923. [PDF] [Code]
  • Hai Wan（万海）, Haicheng Chen, Jianfeng Du, Weilin Luo, Rongzhen Ye: A DQN-based Approach to Finding Precise Evidences for Fact Verification. ACL/IJCNLP (1) 2021: 1030-1039. [PDF] [Code]



---

• 约束求解、组合优化与需求工程 (Constraint Solving, Combinatorial Optimization and Requirements Engineering)

  

  该方向研究布尔可满足性（SAT）与知识编译、基于局部搜索的组合优化，以及面向目标的需求工程，三者以约束求解与组合搜索为共同的方法基础。具体工作包括非子句布尔公式的素蕴涵编译、最小割集计算、改进配置检查策略以缓解局部搜索中的循环问题，以及在目标冲突分析中识别边界条件并加速其可满足性验证。相关成果发表于 ICSE、ICCAD、AAAI 及相关期刊。

  研究论文：
  • Weilin Luo, Polong Chen, Hai Wan（万海）, Hongzhen Zhong, Shaowei Cai, Zhanhao Xiao: Goal-conflict identification based on local search and fast boundary-condition verification based on incremental satisfiability filter. Journal of Systems and Software 213: 112036 (2024). [PDF] [Code]
  • Weilin Luo, Rongzhen Ye, Hai Wan（万海）, Shaowei Cai, Biqing Fang, Delong Zhang: Improving Local Search Algorithms via Probabilistic Configuration Checking. AAAI 2022: 10283-10290. [PDF]
  • Weilin Luo, Hai Wan（万海）, Hongzhen Zhong, Ou Wei, Biqing Fang, Xiaotong Song: An Efficient Two-phase Method for Prime Compilation of Non-clausal Boolean Formulae. ICCAD 2021: 1-9. [PDF] [Code]
  • Weilin Luo, Hai Wan（万海）, Xiaotong Song, Binhao Yang, Hongzhen Zhong, Yin Chen: How to Identify Boundary Conditions with Contrasty Metric? ICSE 2021: 1473-1484. [PDF]
  • Weilin Luo, Ou Wei, Hai Wan（万海）: SATMCS: An Efficient SAT-Based Algorithm and Its Improvements for Computing Minimal Cut Sets. IEEE Transactions on Reliability 70(2): 575-589 (2021).



---

• 机器学习与计算机视觉 (Machine Learning and Computer Vision)

  

  该方向涵盖若干机器学习与计算机视觉方面的研究与合作，方法上以对抗学习和表示学习为主。代表工作包括在无标注数据受污染场景下、借助少量部分观测样本进行的抗污染异常检测，融合结构信息与视觉特征的场景图谱补全，以及基于属性描述的行人重识别。

  研究论文：
  • Wenxi Lv, Qinliang Su, Hai Wan（万海）, Hongteng Xu, Wenchao Xu: Contamination-Resilient Anomaly Detection via Adversarial Learning on Partially-Observed Normal and Anomalous Data. ICML 2024. [PDF] [Code]
  • Hai Wan（万海）, Yonghao Luo, Bo Peng, Wei-Shi Zheng: Representation Learning for Scene Graph Completion via Jointly Structural and Visual Embedding. IJCAI 2018: 949-956.
  • Zhou Yin, Wei-Shi Zheng, Ancong Wu, Hong-Xing Yu, Hai Wan, Xiaowei Guo, Feiyue Huang, Jianhuang Lai. Adversarial Attribute-Image Person Re-identification. IJCAI 2018: 1100-1106.



---

• 知识表示与推理 (Knowledge Representation and Reasoning)

  

  该方向是早期及持续的研究基础，涵盖经典的知识表示与推理课题：多智能体认知规划与高阶信念修正/更新、回答集编程（逻辑程序的环、分裂与化简）、限定（circumscription）、存在规则下的查询回答，以及分层任务网络（HTN）规划与关于动作的推理。这些工作以逻辑为核心，也为上述神经符号方向提供了符号推理的理论支撑。相关成果发表于 Artificial Intelligence、IJCAI、AAAI、ECAI 等。

  研究论文：
  • Hai Wan（万海）, Biqing Fang, Yongmei Liu: A General Multi-agent Epistemic Planner based on Higher-order Belief Change. Artificial Intelligence 301: 103562 (2021). [PDF] [Code]
  • Hai Wan（万海）, Guohui Xiao, Chenglin Wang, Xianqiao Liu, Junhong Chen, Zhe Wang: Query Answering with Guarded Existential Rules under Stable Model Semantics. AAAI 2020: 3017-3024.
  • Zhanhao Xiao, Hai Wan（万海）, Hankz Hankui Zhuo, Andreas Herzig, Laurent Perrussel, Peilin Chen: Refining HTN Methods via Task Insertion with Preferences. AAAI 2020: 10009-10016.
  • Jianmin Ji, Hai Wan, Kewen Wang, Zhe Wang, Chuhan Zhang, Jiangtao Xu: Eliminating Disjunctions in Answer Set Programming by Restricted Unfolding. IJCAI 2016: 1130-1137.
  • Hai Wan（万海）, Zhanhao Xiao, Zhenfeng Yuan, Heng Zhang, Yan Zhang: Computing General First-Order Parallel and Prioritized Circumscription. AAAI 2014: 1105-1111.