A neural network model for temporal sequence learning and motor programming

Abstract

A neural network model for fast learning and storage of temporal sequences is presented. The recall of a learned sequence is triggered by the occurrence of an item relating to its identity, and one of the main distinctive features of this model is that the speed at which a sequence is repeated can be freely modulated by a control subsystem. The possible applications of the model are illustrated by applying it to the production of motor forms. It is shown that any spatial shape memorized in exteroceptive terms can be reproduced in terms of movement by any of the effector systems of the body, and in particular by a simulated jointed arm, at any point in its working space and at any suitable size scale. Our theoretical approach reinforces the idea that the structures responsible for planning a movement in the central nervous system might be largely independent of the motor systems performing this movement.