In the rapidly evolving field of software security, static analysis serves as a cornerstone technique, offering a range of applications such as vulnerability discovery and program hardening. Unfortunately, they suffer from fundamental challenges such as tradeoffs between soundness, precision, and scalability. This talk introduces innovative approaches that significantly enhance the efficacy of static analysis.


First, we explore how Large Language Models (LLMs) can be integrated into static analysis workflows to refine and elevate the analysis process. By leveraging LLMs in a selective and targeted fashion, we aim to enhance the accuracy and contextual understanding of static analysis tools, particularly in complex codebases where traditional methods fall short.


Secondly, the talk will delve into a pioneering hybrid pointer analysis technique that seamlessly unifies data-flow-based and type-based analysis. This novel approach not only consolidates the strengths of both methods but also addresses their individual limitations, leading to a more comprehensive and efficient analysis. This hybrid model promises to scale to large-scale programs such as the Linux kernel and retains significant precision.

作为上海交通大学的杰出校友，钱志云教授目前就职于加州大学河滨分校，他的主要研究兴趣覆盖计算机系统与网络安全的多个主题，特别是如何设计更好的漏洞发现方法、构建更好的分析系统、提出更为自动化的漏洞利用方案等。钱志云教授在计算机安全研究领域已发表超过100篇研究论文，在ACM CCS等旗舰级安全会议上获得杰出论文奖，同时也发现并报告了大量影响现实世界安全的CVE漏洞并获得相关厂商致谢和奖励。除了学术研究领域，钱志云教授的研究工作也为工业界所高度关注，并在Black Hat、Can-SecWest、Off-By-One、Phrack magazine、Linux Kernel Security Summit、Qualcomm SecuritySummit等活动上进行了报告，同时在Pwn20wn以及GeekPwn安全竞赛中收获奖项。