Abstract
Formal methods and machine learning have been successfully applied to detect and classify security threats in a variety of application domains. While both approaches have limitations, we argue for combining the best of them to detect and classify threats in cyberphysical systems.