Abstract
Saccade eye movements are among the most rapid yet precise of all movements produced by higher mammals. Recently we have proposed a spatio-temporal neural network model of the superior colliculus which uses lateral excitatory and inhibitory interconnections to help control both the dynamic and static behavior of saccadic eye movements. In this paper a new learning algorithm integrating genetic algorithms with neural networks for the lateral inhibitory and excitatory interconnections in the saccade generation model is presented. Data base for the training were obtained from neurophysiological experiments, and the training converged well even if random connections were chosen as initial conditions. The resulting network model succeeded in making accurate saccadic eye movements of a variety of sizes while producing realistic spatio-temporal patterns of collicular discharge.
