On error correction in matrix neural networks

摘要

A new model of neural network called a matrix network is proposed. In this network, a distributed record of an initial image is accomplished by choosing synaptic coefficients. Since every neuron of a layer is connected to every neuron of the next layer, the transformation of information in its transmission to the next layer is described by the matrix of such coefficients, where the transmission matrix from the layer of record to the restoration layer is inverse to the transmission matrix from the layer that stores the initial image to the record layer.