Game-Theoretic Algorithms for Optimal Network Security Hardening Using Attack Graphs
Time: Wednesday, December 10th, 2014 at 14:30 in room KN:205.
Penetration testing and optimal network hardening are core problems in network security. For example, a network administrator may need to analyze how best to use limited resources like honeypots to harden a network against possible attacks. Attack graphs are a common formal model used to represent possible attack scenarios for this type of analysis. However, most existing work using attack graphs does not consider the reactions of attackers to different defender strategies. We introduce game-theoretic models that model the joint problem where attacker strategies are represented using attack graphs, and defender strategies are represented as modifications of the attack graph. The attack graphs we use allow for sequential attack actions with associated costs and probabilities of success/failure. Generating an optimal attack plan is an NP-hard subproblem in our model. We translate this planning problem into an MDP and solve it using policy search with several enhancements.