Coevolutionary Approach to Sequential Stackelberg Security Games

Adam Zychowski, Jacek Mandziuk

Abstract: The paper introduces a novel coevolutionary approach (CoEvoSG) for solving Sequential Stackelberg Security Games. CoEvoSG maintains two competing populations of players’ strategies. In the process inspired by biological evolution both populations are developed simultaneously in order to approximate Stackelberg Equilibrium. The comprehensive experimental study based on over 500 test instances of two game types proved CoEvoSG’s ability to repetitively find optimal or close to optimal solutions. The main strength of the proposed method is its time scalability which is highly competitive to the state-of-the-art algorithms and allows to calculate bigger and more complicated games than ever before. Due to the generic and knowledge-free design of CoEvoSG, the method can be applied to diverse real-life security scenarios.