COMBINATORIAL OPTIMIZATION THEORY AND PROBLEM OF ARTIFICIAL INTELLIGENCE

Abstract

The problems of artificial intelligence are complex in nature and not always subject to formalization. But many of the applications of this class are reduced to combinatorial optimization problems. This is due to the fact that their predominant part in finding an optimal solution requires the parsing of options. The parsing problems is characteristic of combinatorial nature. This property can be investigated by simulating the specified problems in the framework of the theory of combinatorial optimization. For their modeling it is necessary to determine the type of these problems (static or dynamic), to determine the argument of the objective function (combinatorial configuration), to modeling the objective function. As the system analysis shows, combinatorial configurations in problems of this class can be both an argument of the objective function and input data. Also, the development of intelligent systems requires the formalization of human natural intelligence, that is, it is necessary to describe the processes of natural thinking and answer the question: can it be simulated. The article gives a classification of natural intelligence, which is developed taking into account the situation of uncertainty. To this end, certain types of uncertainties that arise in solving applied problems of artificial intelligence are considered. The construction of mathematical models of problems of artificial intelligence using the theory of combinatorial optimization is shown on the example of recognition and segmentation of speech signals and clinical diagnostics. It is stated that they are divided into subproblems, which are solved by independent algorithms in the iterative mode. Such a computational scheme is characteristic of hybrid algorithms. By the argument of the objective function the problems of speech recognition and clinical diagnosis – similar to each other. The use of combinatorial optimization theory for modeling the problems of artificial intelligence allows us to establish their combinatorial nature, to formulate the objective function explicitly, to identify the characteristic features that determine the similarity of these problems. The conducted researches allow to reveal the reason of uncertainty of various kinds that arises in the process of their solution, and to explain the nature of the fuzziness of the input data.