Development of the structural interacting technologies scheme for training of neural networks

Abstract

An artificial neural network (NN) is a device of parallel computing, which consists of many interacting processors. Such processors are usually very simple, unlike those used in computers. Each of them works only with signals that it receives and sends at certain intervals. However, when combining such locally simple elements into a sufficiently large network with controlled interaction, it is possible to solve a wide range of quite complex problems. The most important property of NN is the ability to learn based on the processing of environmental data and as a result of learning to increase the level of system performance. Productivity increases over time, according to certain rules. NN training takes place through an interactive process of adjusting synaptic weights and thresholds. At best, NN acquires new knowledge about the environment at each iteration of the learning process. Natural language processing (NLP) is the application of computational methods to model and extract information from human language. With the rise of social media, conversational agents, and personal assistants, computational linguistics is increasingly relevant in creating practical solutions to modeling and understanding human language. In the course of this work, a study of technologies used to teach neural networks without a teacher (unsupervised learning), features and methods of teaching neural networks to solve NLP (Natural Language Processing). Features of natural language processing (NLP) for learning neural networks, methods of morphological, lexical, syntactic, semantic, discourse analysis, methods of classification and clustering of text data, the process of vocabulary formation were considered. Technologies of vector representation of the text, the process of predicate formation (Claim) and its negation (Claim Negation), as well as the process of synthesis of new predicates (Claim Synthesis) are described.