A NEW TYPE OF NEURAL NETWORKS: BAYESIAN MEASUREMENT NEURAL NETWORKS (BIN) BASED ON THE METHODOLOGY OF THE REGULARIZING BAYESIAN APPROACH

UDC 004.032.26

Authors

Prokopchina Svetlana V.
Doctor of Technical Sciences, Professor of the Department «System analysis in Economics», Financial University under the Government of the Russian Federation, Moscow, Russia

Abstract

The paper considers the approach and methodological principles of creating a new type of neural networks, called Bayesian measurement networks. The concept and formalization of a new type of Bayesian neurons implementing Bayesian convolution based on the regularizing Bayesian approach is given. Three types of Bayesian neurons that implement convolution of values of quantitative and qualitative features are considered. An architectural scheme and metrological justification of BIN solutions are proposed. A platform for rapid development of applied BMN is proposed. Examples of solutions to applied problems based on BMN are given

Keywords

Neural network, Bayesian convolution, Regularizing Bayesian approach