Abstract
Bayesian networks are described based on correlation coefficients of the second, third, and the higher orders. The growth of orders for the correlation coefficients is proved to result in significant decrease in probabilities of the 2nd type errors. It testifies indirectly that the efficiency of the multivariate statistical analysis increases if the dimension of the correlation moments goes up. If there is a shift from ordinary correlative composed functions to higher-order composed functions the learning procedure for Bayes-Hamming neural networks does not become complicated.