MULTILAYER MARKOV NETWORKS WITH ABNORMAL INFERENCE

The study is devoted to multilayer Markov networks are not direct output, introduced into scientific circulation in the theory of Markov algorithms. These networks are synthesized from the N-scheme of the Markov algorithm, developed to replace his wellknown γ-scheme. The provisions of the theory of Markov algorithms are developing in the direction of joint processing of words and morphisms. In the schemes of the Markov algorithm γ and N, theoretical differences are considered, which determine the differences in multilayer networks feed-forward with normal Markov output and are not direct output. The theory of Markov algorithms is adopted in conjunction with the theory of categories, which is used to specify words and their compositions, as well as the structure of a Markov network. A definition and a categorical formula for the introduced concept of zero-category are given. The mechanism of synthesis from zero-categories of zero-Markov networks is shown. In the theory of algorithms, the concept of a self-scaling subnet is introduced. For self-priming subnets, theoretical and practical differences from Markov networks are shown. To synthesize a Markov network by a self-scaling subnet, one must be trained in the rules of inference. Self-plating of such a subnet is carried out by the first word from the input data sample coming to its input. Interactions are studied in zero-categories and categories, examples of Markov networks for various compositions of morphisms are given. Possible approaches to weighting morphisms are considered from a theoretical point of view. The mechanism for choosing a layer of a Markov network for weighting morphisms is analyzed. Methods for the formation of compositions of possible inference rules in self-priming subnets are estimated theoretically. It is concluded that the moment of saturation of the Markov network with occurrences, at which it begins to have a stable structure, is of scientific and practical interest. It is expedient to study the self-scaling subnetwork and the nature of its influence on the Markov network precisely at the moment of stabilization of the structure of the latter.