Abstract
A model of neural networks in the vertebrate retina was constructed on the basis of the neurophysiological data, its spatio-temporal characteristics were calculated with a digital computer, and compared them with the psycho-physical data on the human vision. This model consists of three neuronlayers-input (recepter cell) layer, intermediate layer and output (ganglion cell) layer, which are connected each others in series (temporally) and in parallel (spatially), so that receptive fields are constructed. An analysis of these networks was carried out with matrix calculus in parallel connections and Mikusinsky's operational calculus in series connections. Input-output relations for simple spatio-temporal input stimuli were obtained.
The results show quantitatively a good agreement with the fundamental psychophysical data of Mach and Broca-Sulzer phenomena, and those of spatial and temporal frequency responses of the human vision, when a time constant and a threshold level of inhibitory connections are made higher than that of excitatory connections in the intermediate stage.
