RELEVANCY ASSESSMENT FOR ARTIFICIAL NEURAL NETWORKS

摘要

FIELD: physics.;SUBSTANCE: group of inventions relates to computer engineering and can be used in artificial neural networks. Device is configured to redistribute the initial relevance score obtained from the network output to the set of elements by back propagation of the initial relevance score through the artificial neural network in order to obtain a relevance score for each element, wherein the device is configured to perform reverse propagation such that for each neuron, pre-redistributed estimates of relevance of a set of descending neighbouring neurons of the corresponding neuron are distributed over a set of upstream neighbouring neurons of the corresponding neuron using a distribution function.;EFFECT: technical result is to ensure relevance assessment for artificial neural networks.;35 cl, 21 dwg