PhD Proposal by Aris Kanellopoulos
(Advisor: Kyriakos G. Vamvoudakis)
will propose a doctoral thesis entitled,
Control and Game-Theoretic Methods for Secure
Tuesday, May 11 at 11:00 a.m.
Owing to their own intrinsic complexity as well as their exposition to human-centric environments, cyber-physical systems are extremely vulnerable to attacks from adversarial agents, both of machine and human nature. In this work, we will focus on methods that allow for a defending system to gain an advantage in cyber-physical security scenarios by employing tools from control, game and learning theories. Specifically, we will propose a defense approach that brings the principles of Moving Target Defense - resting on the limited ability of an attacker to conduct successful reconnaissance of the system in finite time - into a more rigorous dynamical systems framework. Towards this, we will utilize the system's own complexity, due to sensor and actuator redundancies, and appropriate switching rules, in tandem with a reinforcement learning-inspired detection mechanism. Moreover, we will tackle attacks that operate on the links between systems by proposing a reinforcement learning-based solution to mitigate Denial-of-Service and jamming attacks. The verification problem will also be considered, in order to guarantee that the system will remain operational, despite the presence of structural complexity, through the bridging of reinforcement learning and dissipativity principles. Furthermore, we will take a more abstract view of complex system security by exploring the principles of bounded rationality. We will suggest the use of policy iteration methods to bridge cognitive hierarchy and level-k thinking with dynamical processes. Those ideas, originating from behavioral economics, will enable data-driven attack prediction in a more realistic fashion than what can be offered by game equilibrium solutions. The issue of intelligence in security scenarios will be further considered via the ideas of dynamical deception through a proposed framework where bounded rationality is understood as a hierarchy in learning, rather than optimizing, capability. This viewpoint will allow us to propose methods of exploiting the learning process of an imperfect opponent in order to affect their cognitive state via the use of tools from optimal control theory.
- Prof. Kyriakos G. Vamvoudakis – School of Aerospace Engineering (advisor)
- Prof. Wassim M. Haddad – School of Aerospace Engineering
- Prof. Yorai Wardi – School of Electrical & Computer Engineering
- Prof. João P. Hespanha – Electrical & Computer Engineering Dept., UC Santa Barbara
- Workflow Status: Published
- Created By: Tatianna Richardson
- Created: 04/26/2021
- Modified By: Tatianna Richardson
- Modified: 04/26/2021