Friday, November 10, 2023 | 10:00 AM - 11:00 AM
Preston M. Green Hall, Rodin Auditorium, L0120
135 N Skinker Blvd, St. Louis, MO 63112, USA
Formal Verification of Neural Network-Controlled Autonomous Systems
Abstract: Deep Neural Networks (DNNs) are increasingly used to control physical/mechanical systems. Self-driving cars, drones, and smart cities are just examples of such systems, to name a few. However, regardless of the explosion in the use of DNNs within many cyber-physical systems (CPS) domains, the safety and reliability of these DNN-controlled CPS still need to be investigated. Mathematically based techniques for the specification, development, and verification of software and hardware systems, also known as formal methods, hold the promise to provide appropriate rigorous analysis of the reliability and safety of DNN-controlled CPS. In this talk, I will introduce our tools to verify neural networks against input-output specifications. I will then show how to translate different problems in assured autonomy and certifiable perception into a set of input-output specifications on the neural network controller. Finally, I will extend these results into other domains, including the safe offloading of neural network computations and the fairness of neural network-based decision-making.