Imitation tendencies of local search schemes in baldwinian evolution

Abstract

Baldwinian evolution is a type of hybridization of populationbased global search and individual local search. The individuals take local refining processes, then in selection benefit from the improved fitness, but do not pass on the refined traits the data in to the offspring. The lost information of the refined phenotype implies that the inheritance encoded in genotypes is not directly benefit traits, but the traits having potential to achieve high fitness through the lifetime interaction with the environment. As the result, it is necessary to study how learning works comparing to the previous generation, in addition to how much it improves on the current population. The children may imitate what their parents performed and catch up with them, or alternatively, explore elsewhere and have no idea of where the parents arrived. In this paper, the trade-off is investigated, and it is revealed that in Baldwinian learning, the capability to follow the parents' footprints benefits. With higher imitation tendency, the evolving population can maintain a greater scale of learning potential, and the search results in better speed and convergence.