Evaluation of storage and functioning characteristics of artificial neural networks on the basis of a neurocomputer device

Abstract

Abstract A solution to the problem of evaluating the temporal and quantitative characteristics of the storage and processing of artificial neural networks based on a neurocomputer device is proposed. The most popular and used topologies of artificial neural networks are considered, for which analytical relationships are presented that allow evaluating the training cycle of an artificial neural network, the amount of memory needed and the amount of transmitted data. The proposed relationships differ in that approaches and characteristics unique to neuroprocessor devices are offered for evaluation when implementing the calculations presented in the neural network logical basis.