SaTC: CORE: Small: Data-driven Attack and Defense Modeling for Cyber-physical Systems

Michigan State University, East Lansing MI

Investigators

Abstract

Modern cyber-physical systems (CPS) such as smart buildings or vaccine storage systems involve several connected devices (phones, sensors, and controllers) that perform multiple objectives including temperature or humidity control. However, attacks such as Stuxnet, or the recent Colonial pipeline attack on industrial CPS have exposed major vulnerabilities, demonstrating the inadequacy of current security techniques to guarantee proper functioning of a CPS. An effective response against smart attackers by pro-actively modeling and responding to attacks on a complex CPS at an ongoing basis, especially with limited data on vulnerabilities in the CPS will be developed in this project. Using the concept of hybrid attack graphs to capture discrete and domain-specific (physical) dynamics, a data-driven game-theoretic methodology for securing the cyber and physical components of a CPS is the focus of this project. Three closely integrated research thrusts will span the extent of this project: (1) automated strategies to characterize attacker intent using reinforcement learning (RL); (2) an integrated defense approach to guide the security strategies of the CPS using a novel combination of game theory, RL and Bayesian optimization; and (3) evaluation and validation via realistic simulations with emulated data from sensors and actuators of intelligent buildings. Multiple applications including securing critical infrastructure and supply chains with significant societal impact on public safety will benefit from the project outcomes. Training opportunities for one PhD student and multiple undergraduate students participating in summer research programs, especially from underrepresented groups will be created. Several K-12 outreach and teacher training activities, including hands-on, interactive games and contests between a defender and attacker will be designed to motivate K-12 students to pursue science and engineering. The curriculum of the new graduate course on non-cooperative game theory at MSU will be enriched with new data-driven methods for solving games. A project website has been created (https://github.com/sandeepbanik/Data-Driven-Resilient-Systems) to provide access to all data from this project. Each publication, thesis, or report will have associated with it a set of web pages for the data plots and tables contained therein. In turn, each of these pages will point to files containing the data itself, the metadata and, as relevant, all information needed to regenerate the data. This site will remain active for at least 3 years beyond the end of the award, or the last public release of data associated with the award, whichever is later. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.