Seminar: Engineering Trustworthy Software Change: From Program Analysis to AI Reasoning

_Abstract

Modern software systems are increasingly complex, making it difficult for developers to understand, modify, and secure software without introducing unintended consequences. My research investigates a fundamental software engineering problem: how can we help developers make software changes that are correct, secure, and consistent with their intent? I approach this problem through four closely connected elements—evidence, intent, change, and validation—using techniques including program analysis, program transformation, empirical software engineering, repository mining, and, more recently, AI-based reasoning. 

In this talk, I will present two long-running lines of research that illustrate this vision. First, in secure software development, I will show how our work progresses from understanding developers' secure-coding challenges and evaluating vulnerability-detection tools to learning security repairs from concrete examples and generating behavioral evidence associated with vulnerable dependencies. Second, in software evolution, I will discuss our progression from characterizing and detecting software merge conflicts to inferring developer resolution strategies and generating project-specific conflict resolutions. This work demonstrates how software changes and repository history can provide evidence about developer intent. I will then present our recent work on large language models, which shows that more flexible AI reasoning does not eliminate the need for carefully selected software evidence. 

Finally, I will discuss how these research directions motivate my future vision of AI agents for trustworthy software change. As AI moves from suggesting code toward taking increasingly autonomous actions, I envision agents that can gather relevant software evidence, reason about developer intent, make software changes, and systematically validate their consequences. This vision builds on the central theme of the talk: combining the precision and evidence of software engineering with the flexible reasoning capabilities of modern AI.

_Biography

Na Meng is an Associate Professor in the Department of Computer Science at Virginia Tech since 2021. She received her B.E. in Software Engineering from Northeastern University (NEU) in China in 2006, and received her M.S. in Computer Science from Peking University in China in 2009. She obtained her Ph.D. in Computer Science from The University of Texas at Austin in 2014, advised by Miryung Kim and Kathryn S. McKinley. She started working in Virginia Tech as an Assistant Professor in 2015.

Dr. Meng's research interests span Software Engineering, Programming Languages, Software Security, and Artificial Intelligence. She leads the NiSE (iNnovations in Software Engineering) research group, which conducts various empirical studies and develops novel automated approaches. The group's mission is to uncover new and intriguing phenomena in contemporary software practices, design innovative tools that advance software development as well as maintenance in the future, and create automated solutions that address real-world challenges--ultimately enhancing public well-being and improving human health. Dr. Meng received the NSF CAREER Award in 2019. Her research has been supported by NSF, ONR, CCI, and OpenAI, Infron.ai, and Google.