Autonomous Security Management in Networked Systems through Learning-based Control

Rising Star Symposium on Cyber-Physical Systems Security, Resilience, and Privacy https://shorturl.at/C3RZy Kim Hammar, University of Melbourne https://kim-hammar.com/ Autonomous management of networked systems is required to achieve reliable operation as these systems are becoming increasingly complex and dynamic. In this talk, I will present a general control methodology that enables autonomous management in realistic operating conditions. Conceptually, the methodology comprises three steps: identification of a formal system model, strategy optimization through learning-based control, and validation on a digital twin. Throughout the talk, I will demonstrate how to apply the methodology to different use cases, including incident response, attack recovery, and performance management. Moreover, I will show how to instantiate the methodology with techniques from game theory, reinforcement learning, foundation models, and causality.