Investigations of Wilson's and Traulsen's Group Selection Models in Evolutionary Computation

摘要

Evolving cooperation by evolutionary algorithms is impossible without introducing extra mechanisms. Group selection theory in biology is a good candidate as it explains the evolution of cooperation in nature. Two biological models, Wilson's trait group selection model and Traulsen's group selection model are investigated and compared in evolutionary computation. Three evolutionary algorithms were designed and tested on an n-player prisoner's dilemma problem; two EAs implement the original Wilson and Traulsen models respectively, and one EA extends Traulsen's model. Experimental results show that the latter model introduces high between-group variance, leading to more robustness than the other two in response to parameter changes such as group size, the fraction of cooperators and selection pressure.