Формалізація представлення продукційних правил в Erlang

• Дата: 2020
• Автор: Шаповалова, Світлана Ігорівна
• Формат: Стаття
• Мова: Ukrainian
• Опубліковано: Kamianets-Podilskyi National Ivan Ohiienko University 2020
• Онлайн доступ: http://mcm-tech.kpnu.edu.ua/article/view/216500
• Назва журналу: Mathematical and computer modelling. Series: Technical sciences

The article proposes a method of solving logical puzzles on the basis of machine learning. The method is designed for the preliminary formalization of tasks in the form of description of properties and relations between them. Because each property has a set of possible values, the solution of the puzzle by the methods of search has a combinatorial complexity. With a large number of properties and their values, the time of the solving is rapidly increasing.In recent years, a separate area of research in machine learning has been the solution to logical tasks of this type. However, existing solutions to this area have a number of shortcomings, first and foremost, they do not always guarantee a correct solution.The paper presents a special network of connections for learning the solution of logical puzzles, as well as their formalization for the representation of this network. The network contains computing nodes that represent the relationship between properties, and the nodes of the input layers that specify the values of these relationships.Every task is solved by automatically creating a network of links with its further training until the solution is obtained. The geometric interpretation of the n-dimensional network of bonds and its (n – 1)-dimensional layers is given. The formalization of the presentation of the study sample and the learning algorithm are presented. The mechanism of solving logical combinatorial problems is presented.The article presents examples of tasks that are traditional tests in systems of logical programming and production (expert) systems, as well as tasks from the resource bAbI of such classes: two supporting facts, two argument relations, positional reasoning.The efficiency of the proposed method has been experimentally proved.The prospects of further researches, which are connected with the creation of a lexical-syntactic analyzer for automatic representation of properties, their values and relations between them, are determined.The proposed method is universal and does not depend on the characteristics of the current task, such as the number of properties and their values