BEHAVIORAL PATTERNING BY NEURAL NETWORKS LACKING SENSORV INNERVATION

Abstract

A new approach is offered wherein behavior emitted by neural networks without antecedent stimuli is either shaped to produce a patterned behavioral output (Simulation 1) or is strengthened by delayed reinforcement through the mediation of response afterdischarges (Simulation 2). These networks demonstrate how Stein's InVitro Reinforcement (lVR) of neuronal bursts might account for various reinforcement effects at a behavioral level. The explorations presented iIIustrate two benefits to behavior analysis provided by biobehaviorally-based computational models of learning: accomodation of new biological information and recasting of behavioral concepts in ways compatible with this new information.