Abstract:
Traditionally, program analysis is based on precise, logical reasoning, which typically gets combined with heuristics to ensure that analysis tools are practical. Recent work has shown tremendous success through an alternative way of creating developer tools, which we call neural software analysis. The key idea is to train a neural machine learning model on numerous code examples, which, once trained, makes predictions about previously unseen code. Over the past few years, neural software analysis has evolved from a crazy idea to one of the most active research areas in the field.

This talk discusses recent results and trends in this thriving research area from three perspectives. On the “good” side, the talk will present some neural program analyses that achieve remarkable results – complementing or even outperforming traditional techniques. These analyses address challenging and important developer problems, e.g., by predicting otherwise missing type annotations and by revealing bugs that traditional analyses fail to detect. On the “bad” side, the talk will critically (and perhaps controversially) reflect on difficulties faced by the community, e.g., in obtaining datasets that are large-scale, realistic, and of low noise for tasks that developers indeed need help with. Finally, on the “ugly” side, the talk will discuss the challenge of understanding what neural models of software are actually learning, along with some first steps toward addressing this challenge.

Speaker Biography:
Michael Pradel is a full professor at the University of Stuttgart, which he joined after a PhD at ETH Zurich, a post-doc at UC Berkeley, an assistant professorship at TU Darmstadt, and a sabbatical at Facebook. His research interests span software engineering, programming languages, security, and machine learning, with a focus on tools and techniques for building reliable, efficient, and secure software. In particular, he is interested in dynamic program analysis, test generation, concurrency, performance profiling, JavaScript-based web applications, and machine learning-based program analysis. Michael has been awarded the Software Engineering Award of the Ernst-Denert-Foundation for his dissertation, the Emmy Noether grant by the German Research Foundation (DFG), and an ERC Starting Grant.