CPI Talk - Demystifying and Detecting Bugs in AI Infrastructure Software
CPI would like to extend an invitation to our CPI Talk on Thursday Oct. 3 from 4:00 - 5:30pm in SRADM EC5-1111 Enterprise Theatre , taking place in person.
Speaker: Song Wang - Associate Professor at the Department of Electrical Engineering and Computer Science at York University
CPI Talks are free and open to everyone regardless of affiliation! High school students and non-Waterloo students/staff are also welcome to join.
No prior knowledge will be expected from the audience.
Please register here.
In this CPI Talk, Song Wang will discuss:
Demystifying and Detecting Bugs in AI Infrastructure Software
Abstract:
AI infrastructure software, such as TensorFlow and PyTorch, is essential to powering modern machine learning applications. However, the complexity and scale of these frameworks present significant challenges in ensuring their reliability and security. In this presentation, Song will discuss their recent efforts in characterizing, identifying, and mitigating bugs and vulnerabilities within AI infrastructure software. Song will begin by exploring the characteristics of vulnerabilities in these systems. Next, he will highlight the gaps between current state-of-the-art bug and vulnerability detection methods, such as static analysis and dynamic fuzzing tools, and their effectiveness in detecting AI infrastructure vulnerabilities. Finally, he will introduce their data-driven fuzzing approach for uncovering bugs and vulnerabilities in AI infrastructure software. Additionally, Song will discuss their pilot studies on hunting AI infrastructure software bugs in the context of large language models (LLMs).
Song earned his Ph.D. from the University of Waterloo in January 2019 and then joined York University as an Assistant Professor. He received tenure and was promoted to Associate Professor in 2024. He serves as an Associate Editor for ACM Transactions on Software Engineering and Methodology (TOSEM). His research lies at the intersection of Software Engineering and Artificial Intelligence, with a focus on improving the reliability of AI infrastructure software by characterizing, identifying, and mitigating vulnerabilities, bugs, and failures in frameworks such as TensorFlow and PyTorch. To date, he has published over 60 papers in prestigious Software Engineering journals and flagship conferences. His work has received four best paper awards, including a Distinguished Paper Award at APSEC’23, an ACM Distinguished Paper Award at ICPC’22, another at ICSE’20, and a Best Paper Award at PROMISE’19. Additionally, he was recognized as one of the top 10 most impactful early-career researchers in Software Engineering by the Journal of Systems and Software.