The parallel Nash Memory for asymmetric games

Coevolutionary algorithms search for test cases as part of the search process. The resulting adaptive evaluation function takes away the need to define a fixed evaluation function, but may also be unstable and thereby prevent reliable progress. Recent work in coevolution has therefore focused on algorithms that guarantee progress with respect to a given solution concept. The Nash Memory archive guarantees monotonicity with respect to the game-theoretic solution concept of the Nash equilibrium, but is limited to symmetric games. We present an extension of the Nash Memory that guarantees monotonicity for asymmetric games. The Parallel Nash Memory is demonstrated in experiments, and its performance on general sum games is discussed.