UDC 519.226.3; 004.032.26

Authors

Kozhomberdieva G. I.
Emperor Alexander I St. Petersburg State Transport University, Saint-Petersburg, Russia
Burakov D. P.
Emperor Alexander I St. Petersburg State Transport University, Saint-Petersburg, Russia

Abstract

In the paper, a model of an artificial neuron, the principle of which is based on the application of Bayes' formula is considered. At that, input signals of the neuron are interpreted as evidence in favor of two alternative hypotheses: compliance and non-compliance with the neuron activation condition. The output signal of the neuron is formed on the basis of the posterior probability distribution of hypotheses calculated using the Bayes' formula. An approach to solving the problem of false pattern recognition using the Bayesian model of a neuron is proposed and demonstrated. The features of the application of the Bayesian neuron model in comparison with the traditionally used recognition method are revealed and refined. A method of using the proposed neuron model in neural networks of the WTA type for solving recognition problems is discussed on the example of recognizing the digits of the Russian postal code

Keywords

Artificial neuron, Neural network, Bayes' formula, Posterior probability distribution, Activation function, Bayesian neuron model, Recognition and classification, WTA network, Pattern recognition tasks