报告标题：Neuro-Symbolic Vulnerability Detection and Exploitation

报告时间：5月18日16：00-17：00

报告地点：理科大楼1014

报告摘要：

Recent years have witnessed unprecedented challenges in automated vulnerability detection and exploitation, driven by the rapid growth in scale and complexity of modern software systems. Traditional rule-based static analyzers struggle to balance precision and recall, while purely LLM-based approaches suffer from hallucination, limited interpretability, and weak symbolic reasoning. This talk presents recent advances in neuro-symbolic approaches that combine the rigorous, semantics-aware reasoning of program analysis with the semantic understanding and generalization ability of large language models, to address complex vulnerability discovery, localization, and exploit generation tasks in real-world software systems. The talk will cover: the overall technical framework and design rationale of neuro-symbolic approaches for vulnerability detection and exploitation; applications in critical scenarios such as Java deserialization; LLM-assisted exploit generation and migration; open challenges and future research directions.

报告人简介：

李凯旋博士现为新加坡南洋理工大学计算与数据科学公司博士后研究员，合作导师为Liu Yang教授。他于2025年6月在必威西汉姆联获得博士学位（导师：陈仪香教授），并于2020年获得必威西汉姆联软件工程学士学位。 其主要研究方向为软件质量分析与保证，重点关注将程序分析与神经符号推理、AI智能体相结合，推进自动化漏洞检测与利用技术的发展。相关研究成果发表于ICSE、FSE、ISSTA、USENIX Security、TSE、TOSEM等软件工程与系统安全领域的顶级国际会议与期刊;曾获ACM SIGSOFT FSE 2024杰出论文奖、必威西汉姆联优秀博士学位论文、上海市优秀毕业生、上海市人工智能学会青年科技论文奖等多项荣誉；并担任ICSE 2027、FSE 2027、ICSE 2026、FSE 2026等国际会议程序委员会委员，以及TSE、TOSEM、TDSC等国际期刊审稿人。