Traditional search methods based on sparse vector representations are characterized by high efficiency but limited quality due to their inability to capture semantic relationships in the data. On the other hand, dense vector representations can improve quality by capturing semantic relationships. However, these methods face scalability issues and require significant computational resources. With the development of deep neural networks, including transformer-based architectures, there is a growing interest in combining these two approaches. The purpose of this review paper is to review existing works that use deep models to generate sparse representations.

The use of artificial intelligence systems for critical areas of human life requires trust in the result obtained by the system. An explanation of the obtained solution is important for ensuring trust. Fuzzy systems have the property of explainability due to the presence of a base of production rules in natural language. This work is devoted to the development of a gradient descent algorithm for adjusting the parameters of fuzzy classifiers. Experiments on 38 data sets showed that the use of the developed algorithm for classifiers built by a metaheuristic algorithm statistically significantly increases the accuracy of classification.

The experience of scientific research and practical work has shown that in the field of smart manufacturing systems it is possible and advisable to create mivar expert systems (MES) to increase the intellectualization of decision-making and information processing. This allows to improve the quality of manufactured products and move to a new level of creation of automated control systems for manufacturing systems. To create full-fledged MES in the field of smart manufacturing systems, it is necessary to carry out diverse scientific and practical work, and by people of different specialties, who must be united into a single team. The evolutionary development of each mivar expert system of smart manufacturing systems is ensured by the fact that the MES has the property of evolution and at any time any rule can be added, changed or deleted. This has been repeatedly verified during the implementation of group scientific projects, when new rules for added decision-making and information processing processes for various production systems were evolutionarily added.

At a supersonic velocity of the airflow incident on a channel/intake with curved walls, there is the formation of shock waves, across which flow parameters change abruptly. The location of shock waves can change crucially with small changes in flow conditions at the entrance or exit; moreover, the flow can exhibit non-unique regimes and hysteresis. Transitions between the regimes were studied in previous years for two-dimensional, axisymmetric and 3D convergent-divergent intakes, as well as for bent channels of rectangular cross-section at steady or unsteady changes in the supersonic flow parameters at the entrance. In this paper, we consider a two-dimensional 9%-bent channel with a sufficiently long part located downstream of the bend. The turbulent transonic flow is studied numerically under given pressure fluctuations in the exit section of the channel. Solutions of the system of Reynolds-averaged Navier-Stokes equations, which govern flow parameters and the structure, are obtained with the ANSYS-18.2 software based on the finite element method. The solutions show the occurrence of different regimes at the same steady or oscillatory pressure given at the exit. Conditions for the transition from the regime with multiple reflections of oblique shocks from the walls to the regime with a shock located at the entrance are established. It is shown that the inverse transition cannot be obtained with pressure variations in the exit section of the channel; in order to perform such a transition, one needs an increase in the inflow Mach number or pressure.

This article develops an approach to generating managerial decisions for organizing preventive maintenance of a complex of technical means (CTM) of infocommunication equipment. The decision-making process utilizes the simulation modeling method, enabling the prediction of the need for preventive maintenance and the likelihood of emergencies in CTM operation. The article addresses key challenges associated with the use of simulation modeling for organizing CTM preventive maintenance and proposes solutions to these issues. Particular attention is given to ensuring an optimal response mode to emerging incidents and generating organizational decisions for CTM maintenance based on the unified resource principle. This approach minimizes material and labor costs for conducting preventive maintenance and improves the reliability of ICT.

A simulation program has been developed to assess the financial status of a monetary fund while explicitly accounting for the discrete costs of remediating information-security incidents within a corporate information system. The program’s mathematical core models a fund that first accumulates and then expends the requisite financial resources; the intricate coupling of these inflow and outflow streams renders the underlying stochastic process non-stationary. Simulating this process constitutes the program’s first stage, yielding sample path data from which point and interval estimates of single-factor and two-factor risks are derived. The implementation is written in Python 3.13. Experimental validation produced both scientific insights and practical outcomes. In particular, when the number of payments exceeds one, the distribution – and consequently the magnitude – of those payments markedly influences risk values even for the same payment count. Finally, a function is proposed and verified that approximates the point estimate of the two-factor risk as a function of a coefficient affecting the initial value of the stochastic process.

The quality of the surface of a part formed by the technological process of cutting is assessed by geometric and physical-mechanical parameters. Grinding is used as the final processing operation, in connection with which the accuracy of the obtained surface acquires the greatest importance. The concept of accuracy includes three components: deviation of the shape and location of the surface, waviness and roughness. The article presents the results of a study of the influence of grinding modes on one of the indicators of cutting accuracy – the roughness of the ground surface.

The paper presents a CFD model of a bed reactor for biomass pyrolysis. The input data for the model are kinetic coefficients obtained by an original numerical method based on mathematical processing of thermogravimetric and mass spectrometric curves of biomass pyrolysis. Validation of the model on the basis of experimental results is carried out. The description of the used kinetic model, heat and mass transfer mechanisms is given. The developed model is characterized by coefficients of determination 0.82-0.85 with respect to the experimental data, and shows the best convergence at the temperature program aimed at exposure at 480°C. The developed model can be used for scaling of pyrolysis reactors.

The issues of determining the parameters of the thermal model of Peltier elements necessary for its representation in the computational model of the finite element method are considered. It is shown how, using the reference data of the Peltier element, to obtain an equation approximating its effective thermal power. Using this equation, a method is proposed for determining the effective coefficient of thermal conductivity, volume and surface heat flux density of a Peltier element. These parameters allow you to fully describe the element in a finite element calculation scheme. The obtained values were used to calculate a Peltier element thermostat in the Russian CAE computer-aided design system APM WinMachine.

The objective of the studies presented in the article was to determine the efficiency of using specialized transformers (ST) included in the contact network (CN) section and an additional return wire to reduce electromagnetic pollution at mainline railway facilities. To achieve this objective, computer models of the traction power supply system (TPS) of an AC mainline railway with a CN voltage of 25 kV were developed. The TPS was equipped with ST and additional return wires to enhance the shielding effect of the track in order to reduce electromagnetic field strengths that have a negative impact on personnel, the population and the environment. Multivariate modeling was performed to determine the efficiency of using ST. For comparison, EMF calculations were performed for a typical TPS in which ST were not used. Modeling was carried out in the Fazonord software product developed at IrGUPS and allowing one to calculate TPS modes and determine EMF of the traction network (TN) of various designs. The obtained results showed that installation of specialized transformers of relatively low power of 225 kVA, included in the section of the compressor station and additional return wire, allows to significantly improve the conditions of electromagnetic ecology. This method does not require significant capital expenditures and can be implemented on existing and planned sections of electrified railways. Computer modeling showed that this approach allows to reduce average values of intensity at a height of 1.8 meters for the electric field by 20%, and for the magnetic field - by 166%. Similar indicators for a height of 0.5 m are 20 and 190%, respectively. The physical mechanism for reducing the intensity of EMF consists in enhancing the shielding effect of the track. The approach used in solving problems of electromagnetic ecology is original and differs from known ones in the following provisions: systematicity, consisting in the ability to calculate EMF taking into account the properties and characteristics of a complex traction power system and the supply electric power system; versatility, ensuring modeling of traction networks of various designs; adequacy to the external environment, achieved by taking into account the profile of the underlying surface, underground communications, artificial structures of railway transport, such as galleries, bridges and tunnels; complexity, ensured by combining calculations of the mode and determination of the EMF strengths.

The authors propose a method for simulating daily electrical load schedules in the residential sector based on the theory of random pulses combined with the convolution theory. The objective is to develop a model that takes into account both the random moment of pulse occurrence and its random duration and amplitude. Earlier models based on the convolution of the column-type energy consumption function described by the difference between two Heaviside functions with the probability density of the normal distribution for shifts in the time of switching on household appliances took into account only one random parameter. The fundamental difference of the model proposed in the article is that it takes into account three parameters characterizing each individual electrical load pulse. This allows, when simulating energy consumption of residential buildings, to obtain results that are much closer to the actual values of consumed power, and also to a greater extent correspond to the real dynamics of growth and decline in electricity consumption levels. The authors conduct a simulation modeling of the energy consumption of a small settlement and compare it with the actual energy consumption of a village of comparable size. The feasibility of applying the model to cases with special infrastructural conditions, such as the lack of central heating and water supply, is being tested. Among other things, the possibility of obtaining an optimal modeling result is being tested, taking into account different categories of consumers, such as working and non-working citizens. The possibility of transferring part of the load, which is not typical for night time, to other periods of the day, and the effect of transferring this load on the accuracy of the result are analyzed. Based on the results of the work, the advantages of the developed model are described and the conclusion is made about the possibility of its application in a wide range of tasks.

Various types of representative functions of society management at different levels are analyzed. Their main characteristics are highlighted. A feature of the functioning of agents in the channels of interaction of autonomous systems is the relative independence of their mediation activities, high professional training and specialization. Representative functions are expressed in the representation and protection of the rights and interests of the principal, mainly in the form of coordination, control and operational information. There is a marked alienation of rights and powers in favor of the representative in accordance with the directives, powers of attorney and contracts received. The customer's representatives have high freedom of communication and office work, but they cannot interfere in the work of the operating companies. The representatives improve the interaction of a legal entity with clients or partners, allow them to effectively solve problems in place and time, and ensure their geographical and hierarchical proximity. The operations carried out by the representative are usually secret, taking into account the prevailing conditions, with the choice of ways to rationalize the use of funds and resources in the implementation of directives. The advantage of this management link is that its operation does not require high costs. Conceptual metamodels are constructed in nomination of category theory in the form of commutative diagrams of the management subjects’ connection as arrows of category morphisms. A generalized vector interpretation of the organizational field is given and it is shown that this approach is through, found in various fields of science at different levels of the systems hierarchy. In such metamodel, numerous agents (lawyers, lobbyists, realtors) represent and protect the rights and interests of clients. They fill the space between institutional significant actors and form the organizational and legal field of social action. There are different levels of representation depending on the degree of alienation of the principal’s powers in favor of the executors.

Modern digital technologies allow collecting and systematizing information on the activities of agricultural enterprises during the annual cycle, starting with soil cultivation and ending with harvesting, storage, processing and sale of products with the construction of a plan for a new annual cycle. Access to data from fields, farms, satellites, unmanned aerial vehicles, automated agrometeorological sites contribute to the collection of large volumes of information for managing agricultural production at different stages and planning for the short, medium and long term. The development of data monitoring systems, including the use of precision farming methods and satellite information, expands the possibilities of using mathematical modeling to optimize production in conditions of uncertainty of many production-economic, agroclimatic and environmental characteristics.

The article proposes a comprehensive model for assessing the success of learning based on the methods of recognition theory and processing of multidimensional data within the activity-based direction, which includes a wide range of modern teaching methods based on intelligent analysis of student data, his digital footprint, planning elements in educational and methodological complexes and predicting learning outcomes. The learning trajectory is formed as an element-by-element clustering of interconnected sequences of labor intensity, intermediate results of mastering work programs for a large sample of students. A neural network model for predictive extrapolation of learning outcomes has been developed. The results of experimental studies of the model on data samples accumulated in digital learning management systems are presented.

The article presents an approach to developing a constructor of information knowledge portals based on an ontological description of information. The applied methodology is for constructing knowledge portals based on basic ontologies and ontological design patterns. The constructor of knowledge portals should provide end-to-end technology for creating and filling a specialized Internet resource, as well as a number of intelligent services, such as an ontology editor for modeling the subject area and uploaded resources, a tool for semantic markup of texts based on a ontology and a customizable interface. The peculiarity of the proposed approach is a) the use of basic ontologies of the scientific subject area and resources as a foundation for developing specialized ontologies, b) the use of ontologies to implement the key functionality of the software associated with setting up data visualization, navigation and intelligent search on the target resource, c) putting focus on specialists who do not have experience in programming or web development. This paper proposes a formal model of a constructor of information portals and a model of an information resource created using this designer, and describes important technical solutions to support the end-to-end technology. The created basic components of the technology are demonstrated using the example of creating a simple information system.

This article presents the design and development of a recommender system for selecting services on the geospatial portal of the Institute of system dynamics and control theory SB RAS (IDSCT SB RAS). The system aims to assist researchers in discovering suitable web services by applying advanced data analysis techniques, in particular collaborative filtering. The current challenges in the traditional systems, including the difficulty of service discovery and lack of personalized recommendations, are addressed. A method for recommending web services is proposed, based on an analysis of a user's preferences and their nearest neighbors. The results demonstrate that the proposed system offers an opportunity to find the most relevant services for researchers based on their patterns of use and preferences.

This article presents the results of a study of the factors of managing user attention and improving the perception of visual information on highly loaded educational portals. The design of information websites has a significant impact on the efficiency of information transmission within the framework of content overload. The results served as the basis for changing the stimulus material in order to identify statistical patterns. The experiment was conducted using eye tracking technology and included 16 stimuli and 6 tasks. As a result of the analysis of variance, the influence of the combination of color scheme and background text selection on the speed of content perception was revealed. The speed of perception is a criterion for evaluating the effectiveness of human-computer interaction and data interpretation. The above research can be the basis for further experiments and the development of recommendations for the design of effective information web resources.