Safety Verification via Deep Learning

Speaker Info

Professor Baruah joined Washington University in St. Louis in September 2017. He was previously at the University of North Carolina at Chapel Hill (1999-2017) and the University of Vermont (1993-1999). His research interests and activities are in real-time and safety-critical system design, scheduling theory, resource allocation and sharing in distributed computing environments, and algorithm design and analysis. He is a Fellow of the IEEE, and the recipient of the 2014 Outstanding Technical Contributions and Leadership Award of the IEEE Technical Committee on Real-Time Systems.

Abstract

Many modern safety-critical cyber-physical systems are characterized by highly dynamic workloads that repeatedly evolve as the system executes, thereby requiring that safety properties be repeatedly re-verified during run-time. Verification of many important safety properties is computationally highly intractable; it is therefore tempting to use Deep-Learning (DL) based techniques to classify system specifications according to whether they do or do not possess relevant safety properties. But how does one do so in a manner that does not compromise system safety despite the well-known fact that DL-based classification is prone to occasional errors? — this issue will be investigated in this presentation, with a particular focus on the verification of safety properties relating to timing correctness (i.e., schedulability).