MODELING THE MAINTENANCE OF HORIZONTAL PURSUITS WITH NEURAL NETWORKS .2. TORQUE CALCULATION NETWORK

A torque calculation network is discussed as the second main part of t he presented computational model of horizontal smooth pursuit eye move ments of primates. In the model, simulated target motion is detected b y a motion detector circuit whose output together with the present eye orientation is sent to this three-layer neural network trained with i mproved back propagation learning paradigm. A torque that activates th e muscles to rotate eyes is generated from the output layer of the net work so as to keep the target at the fovea. Results of simulation show that the trained model can well pursuit a little target which is movi ng horizontally in the visual field with background noise no matter wh ether the velocity of the target is constant or not. Moreover, propert ies inside the model showed some similarities with primate visual cort ex.