Title: Learning, Games, and Control for Security of Cyber-physical Systems
58th Conference on Decision and Control – Nice, France – December 11th-13th 2019
Time and Location: Full day, Dec. 10, 8:30-17:30 – Méditerranée 1
Organizers: Quanyan Zhu (New York University) and Radha Poovendran (University of Washington)
Workshop Description:
The topic of this workshop is the control and secure operation of cyber-physical systems (CPSs) using perspectives from game theory and machine learning. Cyber-physical systems are complex entities where the working of a physical system is governed by its interactions with computing devices and algorithms. These systems are ubiquitous. Examples include medical devices and robots on a small scale, to power systems and connected communities on a large scale. CPSs are expected to operate in dynamically changing environments, which could result in it being the target of malicious attacks that aim to prevent it from accomplishing a goal. Strategies to mitigate the effect of an attack must take into consideration the fact that adversaries are often stealthy, intelligent, and persistent.
Game theory (GT) models scenarios when there are multiple agents who seek to influence the working of the system, and these agents have competing objectives. With the increasing availability of data pertaining to the operation of CPSs (e.g. system log data, user surveys), machine learning (ML) is an increasingly popular solution approach that enables analysis in the absence of a model of the system, and a data-driven approach for the synthesis of control policies for these systems.
The proposed workshop will present a holistic framework for CPS security from multiple perspectives. GT will be used to design incentive mechanisms, and model adversary behavior, including the behavior of human adversaries. When limited information is available about the adversary’s abilities, algorithms can be devised in order to aid the agents in learning optimal security policies. ML will also be useful in developing agent-strategies when there is more than one intelligent attacker. Formal methods (FM) can be used to specify additional constraints of the environment that the agent must satisfy in order to reach a goal.
This workshop will feature talks by leading experts whose recent work uses game theory and data-driven approaches to model and analyze the security of CPSs. The workshop features a presentation by representatives from a funding agency, and a panel discussion in order to identify open research problems that will be of interest to the broader community.
Tentative Schedule:
8:30-8:40am: Opening remarks (Organizers)
8:40-9:20am: Tamer Basar (UIUC) Talk Title: An introduction to games, incentives, and learning in the context of CPS security
9:20-10:00am: Joao Hespanha (UCSB) Talk Title: Games of Partial Information in Cyber-Physical Systems
10:00-10:30am: Morning coffee break
10:30-11:10am: Karl Johannsson (KTH) Talk Title: Cyber Security and Privacy in Networked Control Systems
11:10-11:50am: Linda Bushnell (University of Washington) Talk Title: Vehicle Network Security and Resiliency: Analysis of the Controller Area Network Protocol
11:50-1:30pm: Lunch
1:30-2:10pm: Radha Poovendran (University of Washington) Talk Title: Dynamic Information Flow Tracking Games for APT Detection.
2:10-2:50pm: Quanyan Zhu (NYU) Talk Title: Entanglement of Learning and Security
2:50-3:20pm: Affternoon coffee break
3:20-4:00pm: Hideaki Ishii (Tokyo Institute of Technology) Talk Title: Randomized Communication Protocols for Secure Networked Control Under Jamming Attacks
4:00-5:30pm: Government and Academia Panel
(Including: Jonathan Sprinkle (NSF), Sandip Roy (NSF), Gary Toth (ONR), Ryan Craven (ONR), Fred Leve (AFSOR))
5:30-5:35pm: Closing remarks and discussions