Tournament Competition and its Merits for Coevolutionary Algorithms

Abstract

In a competitive coevolutionary algorithm, the competition strategy, selecting opposing individuals, has a great influence on the performance of the algorithm. Therefore, a good competition strategy is crucial for an effective and efficient competitive coevolutionary algorithm. In this paper, we propose a new competition strategy called tournament competition. We investigate its characteristics and merits when applied to adversarial problems. To verify the performance of the new strategy, we first classify adversarial problems into two types: solution-test problems and game problems. We apply the strategy to both types. A set of experiments compares our strategy to several existing competition strategies and analyzes several aspects such as solution quality, evolution speed, and coevolutionary balance. The experimental results indicate that some of the existing competition strategies give different patterns according to problem types. The results also support that the proposed tournament strategy has the potential for finding good solutions, regardless of problem types.