Heterogeneity and specialization in evolving teams
Terence Soule · 2000
Cooperative teams can be used to improve the performance of standard genetic programming. However, designing appropriate team member choice and cooperation mechanisms for producing successful teams is an important and sometimes di cult task. This task is complicated by the type of programs being generated. For example, programs producing boolean valued outputs can use simple voting as a cooperation mechanism whereas programs with real valued outputs cannot. This research examines several team choice and cooperation mechanisms for linear regression problem. The results suggests that cooperation produces the most improvement when the mechanisms favors specialization. The best results are achieved when each team member can focus on a relatively small portion of the problem domain and when the member's inuence outside of that domain is relatively small. Specialization is found to be a function of both the team choice and cooperation mechanism. 1