NEURAL NETWORK-BASED NEURAL NETWORK CONSTRUCTION WITH FUNCTIONAL TRANSFORMATION OF ARBITRARY TYPE FOR INPUT SIGNALS

摘要

FIELD: computer systems.SUBSTANCE: invention can be used for constructing neural networks for temporal or spatial prediction of parameters of studied objects or pattern recognition. Computer system is realized based on neural network, which includes at least one neuron, using functional transformation of arbitrary type, specified in the form of a table (array) of matching pairs of input data and an output function for each input signal with subsequent summation, wherein values of tabular transformation functions are selected stochastically based on a genetic algorithm when training a neural network.EFFECT: technical result is higher degree of freedom of individual neurons and thereby reduced depth of network.1 cl, 5 dwg