Global motion discrimination using more physiological modified artificial neural networks

Unsupervised neural networks (NNs) have been used to successfully simulate psycho-physical results of learning direction discrimination in global motion. This paper uses an NN with classical back propagation to implement supervised learning as a vehicle to simulate certain psycho-physical and physiological processes. The two most important concepts dealt with are the noise within the neurons and the use of an integrate and fire method of transmission from cell to cell. Each of these `physiological´ additions to the NN model was examined with respect to its effect on network error progression and network robustness in the presence of stimulus noise as well as intrinsic neural noise