Reinforcement Learning Produces Dominant Strategies for the Iterated Prisoner's Dilemma

摘要

We present tournament results and several powerful strategies for theIterated Prisoner's Dilemma created using reinforcement learning techniques(evolutionary and particle swarm algorithms). These strategies are trained toperform well against a corpus of over 170 distinct opponents, including manywell-known and classic strategies. All the trained strategies win standardtournaments against the total collection of other opponents. The trainedstrategies and one particular human made designed strategy are the topperformers in noisy tournaments also.