Turing Learning with Nash Memory
Shuai Wang

Abstract:
Turing Learning is a method for the reverse engineering of agent behaviors. This approach was inspired by the Turing test where a machine can pass if its behaviour is indistinguishable from that of a human. Nash memory is a memory mechanism for coevolution. It guarantees monotonicity in convergence. This thesis explores the integration of such memory mechanism with Turing Learning for faster learning of agent behaviors. We employ the Enki robot simulation platform and learns the aggregation behavior of epuck robots. Our experiments indicate that using Nash memory can reduce the computation time by 35.4% and results in faster convergence for the aggregation game. In addition, we present TuringLearner, the first Turing Learning platform. Keywords: Turing Learning, Nash Memory, Game Theory, Multi-agent System.