Saccadic eye-movement system modelling using recurrent neural network

Saccadic eye-movement is a rapid eye-movement caused by visual or auditory stimulus. There are more than a few models of this system due to relatively abundant experimental data. But existing models are mostly block-diagram models which cannot account for the behavior of neurons at the level of individual neurons, which might explain such important properties for saccades as vector averaging. In this research a connectionist model for the saccadic eye-movement generation system was constructed. It includes signal path ways from superior colliculus, where saccade signal is first initiated, to motor neurons. A recurrent neural network which has various feedforward and feedback connections was trained by using the continuous recurrent backpropagation method. Then the result was compared with neurophysiological data. The result shows that the proposed model does reproduce vector averaging property as well as some individual neuron model in hidden layers exhibit quantitatively similar behavior to recorded neurons.