Genetic approximations for the failure-free security games

Abstract

This paper deals with computational aspects of attack trees, more precisely, evaluating the expected adversarial utility in the failurefree game, where the adversary is allowed to re-run failed atomic attacks an unlimited number of times. It has been shown by Buldas and Lenin that exact evaluation of this utility is an NP-complete problem, so a computationally feasible approximation is needed. In this paper we consider a genetic approach for this challenge. Since genetic algorithms depend on a number of non-trivial parameters, we face a multi-objective optimization problem and we consider several heuristic criteria to solve it.