Informacionnye Tehnologii

The purpose of this article is to develop principles for assessing the similarity between multiparameter objects whose states are described by disparate characteristics. The parameters of many objects are described using different metrological scales (names, ranks, intervals, ratios). These parameters may be described by different units of measurement and have different levels of importance for the degree of similarity between pairs of objects. When comparing the parameters of a test object, there may be various permissible deviations, either upward or downward, relative to the reference object. Furthermore, there may be other conditions under which a test object can be considered similar to the reference object. An analysis of metrics for determining closeness between objects, as well as similarity measures, diversity measures, and inclusion measures, revealed that these theoretical principles have limitations when determining similarity between multiparameter objects. To address these limitations, we propose theoretical principles based on the formation of conditions that allow for recognizing objects as similar for each parameter, as well as additional conditions. Recognition of similarity for specific parameters between a test object and a reference object is accomplished by defining a range of permissible values. If the parameter of the compared object falls within the acceptable range, a decision is generated as one; if it does not, a decision is generated as zero. As a result, a sequence consisting of zeros and ones is formed for the compared object. The elements of the sequence are then summed, taking into account the weighting coefficient of each parameter. The values of the weighting coefficients are distributed so that their sum equals one. As a result, the closer the sum of the sequence elements is to one, the greater the similarity between the objects is considered. To assess the feasibility of the proposed provisions, a sample of values describing the condition of forest areas was used. The results demonstrate the formation of subgroups of objects whose parameters fall within acceptable value ranges and satisfy an additional condition relative to the reference object. The proposed provisions open up opportunities for the further development of information systems for analyzing the condition of complex objects.

This paper concerns the use of Deep Learning methods and Neural Networks for the task of segmentation and classification of hydrophysical time series in order to identify anomalies of a certain type caused by hardware failures. Two models based on Bidirectional Long-Short Term Memory networks and one based on a Temporal Convolutional Network were used. We made the performance comparisons and assessed the detection success. The results obtained demonstrate the effectiveness of the proposed approaches, and it was found that time-frequency analysis significantly improves the models’ ability to detect anomalies of different types.

The author has developed a system for multi-criteria assessment of the quality of application software user interfaces based on artificial intelligence methods. The proposed system is based on a fuzzy expert model that implements the Mamdani inference mechanism, which provides an objective assessment of the user interface according to key criteria: accessibility, adaptability, functionality and navigation structure. Based on the analysis of input parameters, the system determines the final level of interface quality and generates recommendations for improving the evaluated interfaces. The results of the study may be of interest to specialists involved in the development, design and evaluation of user interfaces in various subject areas.

A model of reactive data management in distributed systems with client-server architecture based on finite state machines is proposed. Methods for ensuring the relevance and consistency of data, as well as eliminating duplication of requests and concurrency modes are described. А mathematical model of a data set is developed, including the formalization of collections of key-value pairs. An approach to reactive management through a finite state machine is implemented, ensuring synchronization of data states between the client and the server. It is shown that the proposed model reduces the load on the network and computing resources of the client by optimizing caching and conflict handling.

A new linguistic language model is proposed for solving the problems of analytical and synthetic processing of textual information. Its distinctive feature is the use of the system of inflectional classes of the Russian language, the principle of linguistic analogy, the centroid-contextual model and a wide range of submodels developed by the authors to implement the processes of semantic, syntactic and conceptual analysis of texts. This approach provides an opportunity to create a universal industry language model, which can be used to implement artificial intelligence technologies in a number of high-tech industries. It provides a high degree of automation of technological processes for creating and adapting model components and is based on a modern theoretical concept of understanding the semantic structure of natural language texts, as well as using a set of interrelated linguistic submodels with an extensive feature space developed by the authors. The proposed methods and models of semantic analysis are based on a new approach to multi-level semantic analysis of texts and make it possible to significantly expand the range of domestic technologies for processing Russian-language texts of scientific and technical information. Therefore, it is advisable to use them in the information support system for the implementation of the new strategy of scientific and technological development of Russia.