Automatic Acquisition of Strategies by Co-evolutionary Learning

Co-evolutionary learning is a new learning approach emerged in recent years. It is designed to deal with dynamic learning tasks, i.e., the target to be learned changes over time. Such learning problems are particularly diicult to solve by traditional machine learning methods. This paper studies co-evolutionary learning, identiies its strength and weakness, and presents some experimental results using the N-player Iterated Prisoner's Dilemma game (NIPD). The paper also investigates the impact of group size N on the evolution of strategies and generalisation issue in co-evolutionary learning.