活动预告｜CodeWisdom 软件智能化开发与运维系列学术报告第17期：彻底改变大规模系统的质量保证-CSDN博客

Chenyuan Yang

PhD student at the University of Illinois Urbana-Champaign (UIUC)

Summary

Title

The LLM Bug Hunters: Revolutionizing Quality Assurance for Large-Scale Systems

Abstract

Large Language Models (LLMs) are revolutionizing quality assurance for large-scale systems. This talk will present a journey of innovative LLM-based approaches that progressively enhance software testing and analysis. We begin with TitanFuzz, the first LLM-powered fuzzer, demonstrating foundational test case generation for real-world systems. Building on this, WhiteFox showcases more guided generation, using LLMs to analyze compiler optimization logic for targeted white-box fuzzing. We then explore deeper integration with KernelGPT, which incorporates LLMs into existing fuzzing frameworks to significantly improve Linux kernel fuzzing—the first LLM application in this critical domain. Finally, we extend beyond fuzzing to static analysis with KNighter, an innovative method where LLMs synthesize precise static analysis checkers from past bug patterns, rather than directly analyzing code. These LLM-driven methodologies have proven highly effective, uncovering over 160 critical bugs in machine learning systems (e.g., PyTorch and TensorFlow) and more than 110 bugs in the Linux kernel, with 25 of these recognized as CVEs. These findings underscore the significant potential of LLMs to strengthen the complex software systems essential to our digital infrastructure.

Speaker

Chenyuan Yang is a third-year PhD student at the University of Illinois Urbana-Champaign (UIUC), advised by Prof. Lingming Zhang. His research focuses on the intersection of software systems and machine learning, aiming to improve the reliability of large-scale systems. He achieves this by leveraging and optimizing Large Language Models (LLMs) with testing, reasoning, and verification techniques. His research has led to the detection of over 650 critical bugs in ML systems, C/C++ compilers, and operating systems, including 25 CVEs.

His work has been published in top-tier conferences spanning software engineering, programming languages, machine learning, and computer systems, including five first-author papers. He holds a B.S. degree with honors from the elite program in the Department of Computer Science at Nanjing University, advised by Prof. Yanyan Jiang. His personal website is available at https://yangchenyuan.github.io/.