ON NEURON MATRICES

Neuron matrices and neuron operators are introduced for the description and analysis of multilayer neuron determined networks with particular reference to the visual analyser neurons. It is assumed that every layer is composed of neurons of only one type. It is shown that the addition and subtraction operations of common matrices are specific cases of neuron matrices. The concept of neuron matrices allows one to construct a model of the peripheral part of the visual analyser on the principle of physiological funnels, to distinguish between convex and non‐convex images. The solution of the six Rosenblatt problems concerning the local detectors of the visual analyser is considered.