The self-organization of the orientation selectivity through learning by inhibitory interneurons

A three-layered model of a neural network is proposed, where the learning by interneurons plays an important role in self-organizing the orientation selectivity. The first layer simply provides input patterns to the 2nd layer through plastic excitatory synapses and also to the 3rd layer through constant excitatory synapses. Each neuron of the 2nd layer gives inhibitory output to the corresponding neuron of the 3rd layer. Lateral inhibitions are also assumed among the neurons of the 3rd layers. The bar patterns of various orientation are applied as input to the neural network. After applying the input patterns for a long time, the synaptic strengths cease to be modified and come to an equilibrium. We carry out numerical calculations for the equilibrium and obtained the orientation selectivity in the 3rd layer of the output neurons. The numerical results are shown and discussed.