Many fundamental areas of modern life depend on the flawless functioning of software systems: These systems control and structure operational processes, production lines, transport systems, the supply of electricity, water, food, and much more. However, they are prone to programming errors that can lead to program crashes, malfunctions, or security vulnerabilities. Michael Pradel’s research on software testing and analysis tends to be application-oriented and helps developers avoid errors when programming software. In his Software Lab, Pradel develops prototypes of tools and actively seeks interaction with the developer community.
“We want to give developers the best possible tools to work with”
The Software Lab, Michael Pradel‘s eleven-member research group at the University of Stuttgart, conduct their research in the field of Software Engineering, focusing mainly on questions of software quality. This involves security as much as reliability and efficiency. For Pradel and his group, foundational research and application-oriented research go hand in hand, as he explains: “We conduct foundational research, but we also conduct research that is very close to what developers do every day. Usually, we tend to be a few years ahead of the big companies – and in this sense, it is foundational research – but we aren‘t a long way away from practical application either.”
To establish dialogue with practitioners, Pradel collaborates with open-source developers and maintains industry contacts. In 2019, for instance, he spent a sabbatical at Facebook in California to put ideas from research into practice. About the work of his Software Lab, he says: “We build tools for software developers to optimize the quality of software: Programs that developers can use to write their programs and their code – we want to give them the best possible tools to work with.”
“Pioneering work at the intersection of AI and software”
Traditional developer tools that are used for the analysis of software programs are based on purely mathematical, logical reasoning. The analyses yielded by these tools are often incomplete, because software is after all made by humans. This means that software is not exclusively structured in a way that is precise, logical and efficient, but is also based on conventions and commonly used implementations of a programming language. Like natural languages, programming languages too can be used more or less idiomatically. This information remains hidden from conventional tools, as Pradel explains: “Techniques based on symbolic logic view programs as mathematical formulas that they go through step by step in order to see what the program is actually doing. What they don’t see, unfortunately, is what the program was intended to be doing.”
To close this gap, Pradel intends to further integrate these conventional techniques with an AI-based, neural approach. Not only do the big, elaborately trained LLMs from third-party providers possess extensive knowledge of human thought and behavior, they are also familiar with programs and code. This knowledge can be harnessed for software analysis. “The neural approaches”, says Pradel, “are better able to look at software like a human would and can better understand what the program is intended to do. If we push forward the integration of techniques based on symbolic logic and techniques based on AI, getting the best of both approaches, we are better equipped to address the challenges of software development.”
“Excellent research that has impact in the real world”
With his Software Lab, Pradel was one of the first researchers in the field of software engineering to have pursued and promoted this neuro-symbolic approach. “We began with this mixture of AI and symbolic-logic techniques, which do not involve AI, relatively early, about nine years ago, becoming the pioneers in this area at the intersection of AI and software”, Pradel says. With the resources that will be available to him as CISPA-Faculty, he wants to develop increasingly useful tools: “The latest thing that we want to build are autonomous agents who you can set a task – for example, find vulnerabilities in this software and fix them – and who go on to solve this task more or less independently by using tools themselves and imitating what a human would do in this situation.” Why he thinks CISPA the right place for his pioneering work is easy for Pradel to explain: „The focus at CISPA is where I would see the focus of my work, too: Conducting excellent research that hopefully has impact in the real world.”
About Michael Pradel
Professor Dr. Michael Pradel’s research focuses on software engineering, programming languages, IT security and machine learning. He is professor of computer science at the University of Stuttgart, where he also leads the Software Lab. Previously, Pradel was a postdoctoral researcher at both ETH Zurich and the University of California, Berkeley, an assistant professor at TU Darmstadt, and he spent a sabbatical at Facebook in Menlo Park, California. In 2012, he completed his PhD on “Program Analyses for Automatic and Precise Error Detection” at ETH Zurich. Pradel studied computer science at TU Dresden and engineering at École Centrale Paris.
Michael Pradel’s research has been recognized with numerous international awards. Among these are Ernst-Denert Software Engineering Award, an ERC Starting Grant as well as an ERC Proof of Concept Grant. In 2022, the Association for Computing Machinery (ACM) named him a Distinguished Member, honoring his excellent research results, which are already being used in software developer tools. Pradel’s conference contributions have been honored with, among others, five ACM SIGSOFT Distinguished Paper Awards and two ACM SIGSOFT Distinguished Artifacts Awards.
