Perception redux: Emergence of structure

Abstract

Perceptions were evolved that computed a rather difficult nonlinear Boolean function. The results with this early and basic form of emergent computation suggested that when genetic search is applied to its structure, a perceptron can learn more complex tasks than is sometimes supposed. The results also suggested, in the light of related work on classifier systems, that to hasten the emergence of an emergent computation it is desirable to provide evaluative feedback at a level as close as possible to that of the constituent local computations.