The Effects of Learning on the Roles of Chance, History and Adaptation in Evolving Neural Networks

Abstract

The course of the evolution of a population is affected by chance events, the population’s genetic history and adaptation via selection. The presence of individual lifetime learning is also known to influence the course of a population’s evolution. The experiments reported here, examine the effects that lifetime learning has on the roles played by chance, history and adaptation in the evolution of populations of simple neural networks. The effects of chance, history and adaptation on both learned fitness (fitness after learning) and innate fitness were considered both when learning incurred no cost and when a fitness cost was incurred for learning. When learning was cost-free it was found to decrease the influence of adaptation, history and chance on learned fitness, while having the opposite or possibly no effect on innate fitness. When a fitness cost was incurred for learning, the role of adaptation in determining innate fitness increased, while the roles of chance and history decreased for both learned and innate fitness. These observed effects are interpreted in light prior results on the effects of learning on evolution.