Proposal of probabilistically and dynamically separating GA

摘要

We propose a "probabilistically and dynamically-separating genetic algorithm (pDS-GA)" that is applied to a multi-agent system (MAS). The proposed pDS-GA holds two advantages over a conventional DS-GA: (1) evolution of cooperation can be realized without using a gene network. Therefore, the communication cost between agents as well as the calculation cost will be mitigated. (2) Dynamic separation is based on probability. Therefore, there is no need to always grasp the number of agents in a colony. We verified the character of pDS-GA experimentally, finding that organization by division of work could be realized as a result of the evolution of cooperation.