Impacts of Game-Theoretic and Behavioral Decision-Making on the Robustness and Security of Shared Systems and Networks
We investigate the impacts of game-theoretic and behavioral decision-making in two broad classes of problems: i) resource sharing games, and ii) network security games. In the first part of the thesis, we consider a set of decision-makers or players who choose their levels of utilization of a shared resource. The return from the resource depends on the utilization by all players, and the resource is prone to failure due to overutilization. This problem has been studied in a multitude of disciplines, including engineering (to model congestion in transportation and communication networks) and economics (to model competition over open-access natural resources). We provide a mathematically rigorous characterization of the Nash equilibrium when the players have behavioral risk preferences. Specifically, we model the risk preferences of the players according to prospect theory, a widely accepted and empirically grounded behavioral model of human decision-making under risk and uncertainty. Our analysis quantifies the increase in resource utilization and failure probability at equilibrium due to higher competition and heterogeneity in the risk preferences of the players for a broad class of resource characteristics. We then investigate how behavioral players respond to economic incentives such as taxes imposed by a central planner to control the utilization of the resource. In the second part of the thesis, we consider the impacts of game-theoretic and behavioral decision-making on the security of networked systems. While networks capture relationships and interdependencies in many socio-technical systems, these interconnections often expose the entities/nodes to different types of security risks. We first investigate the impacts of nonlinear (prospect-theoretic) perception of attack/infection probabilities on the security investments at the Nash equilibria in two game-theoretic settings. Specifically, we consider i) interdependent security games, where the attack probability faced by a node depends on the decisions made by her immediate neighbors, and ii) SIS epidemics on networks, where the infection probability of a node depends on the decisions made by all nodes in the network. In both settings, we identify conditions under which nonlinear perception of probabilities lead to improved security outcomes at the respective Nash equilibria. We further characterize the structure of networks that minimize bounds on the expected fraction of nodes that are i) attacked at the equilibria in a class of interdependent security games, and ii) infected in the endemic state of SIS epidemic dynamics. Finally, we propose a game-theoretic framework to investigate the implications of decentralized defense strategies in large-scale networks against targeted attacks. In total, our investigation leads to new insights into the impacts of behavioral and game-theoretic decision-making on the security and robustness of shared systems and networks.
Sundaram, Purdue University.
Electrical engineering|Systems science|Operations research
Off-Campus Purdue Users:
To access this dissertation, please log in to our