Connectionist models of cortico-basal ganglia adaptive neural networks during learning of motor sequential procedures

Abstract

In this paper two neural models of basal ganglia function during motor sequential behaviour are presented. Two connectionist models of neuron - like elements that mimic some aspects of anatomy and physiology of cortico - basal ganglia - thalamo - cortical loops have been developed. The aim of this work is to report a new computational model of motor sequence learning guided by reinforcement signals from neuronal systems that evaluate behaviours. The models are partially recurrent neural networks known as Jordan networks trained under a reinforcement learning paradigm. To validate these models, experimental findings of Tanji and Shima [5] on monkeys have been reviewed. The hypothesis that cortico - basal ganglionic loops learn and perform sequences successfully driven by reinforcement signals has been demonstrated in computer simulations of the models presented in this paper. © Springer-Verlag Berlin Heidelberg 2001.