Modification of K2 algorithm for adaptation of Bayesian neural network

Abstract

The problem of Bayesian neural network adaptation to new data was formulated and analyzed in accordance with the Bayesian approach to obtain a set of possible strategies for the problem solution. For concrete strategy, an algorithm for incremental adaptation of the Bayesian neural network to new observations was developed. The К2 network construction batch algorithm was used as a basis of the adaptation algorithm to provide necessary adaptation criteria.