Meeting room 206 (IMAG)
17 April 2018 - 11h00
Verification of Learning-Enabled, Cyber-Physical Systems
by Jim Kapinski from TEMA Toyota, USA
Abstract: Cyber-physical systems (CPSs) are used in many mission critical applications, such as automobiles, aircraft, and medical devices; therefore, it is vital to ensure that these systems behave correctly. Designs for modern CPSs often include learning-enabled (LE) components, such as neural networks and support vector machines, whose behaviors emerge as a result of processing training data; however, verifying correct behavior of LE aspects of CPS designs is difficult, as the training data and resulting system behaviors can be hard to characterize. To address these challenges, new analysis techniques are needed to increase confidence in the reliability of LE CPSs. This talk describes ongoing challenges in verification for LE CPSs and presents new methods to address these challenges, including best-effort methods that use testing-based approaches and formal methods that use automated reasoning tools to generate proofs of correctness.