DOI: 10.26102/2310-6018/2024.46.3.016
Research in the field of large language models and natural language processing systems has intensified due to the emergence of new, latent and serious risks, for example, violations of the output generation processes, malicious requests in automatic mode. Synergistic scenarios for large language models are being developed. The main hypothesis taken into account in this study is the possibility of insurance (with a given probability) from the generation of prohibited content and its "mixing" with the user query, taking into account ontological properties and connections to improve the quality of search in practical tasks, for example, using an ontology library. Methods of analysis-synthesis, modeling-forecasting, expert-heuristic, probability theory and decision-making were used. The main results of the article: 1) analytics on the problems of applying large language models in achieving stability in the system infrastructure (a table of key methods was proposed); 2) a language model of network infrastructure stability based on estimates of distributions when mixing words is proposed, which uses the Bayesian method; 3) a similar language model was proposed and studied on the basis of an expert-heuristic approach to assessing risks (uncertainties in the system), in particular, using an information-entropy approach. Research can be developed by complicating models (hypotheses) and the "depth" of risk accounting.
Keywords: large language models, resilience, risks, information security, governance
DOI: 10.26102/2310-6018/2024.46.3.018
Hose cable is one of the key management tools, for example in a subsea oil and gas production system. It can be considered as a customized product related to specific parameters of use cases, such as installation location. This paper applies a method to calculate the reliability of the hose cable using the Advanced First Order Second Moment Method (AFOSM) and Monte Carlo method. The advantages and current limitations of adopting a knowledge-based engineering (KBE) approach are discussed, which in turn enables the creation of different product configurations and variants, for the integration of CAD models augmented with an automatic calculation function. Recommendations are made for future research into the KBE method of product design. The paper demonstrates the use of Siemens NX and its framework for representing engineering knowledge called Knowledge Fusion (KF) to create a reliability-aware parametric model of a hose cable design to improve the sectional design process. The benefits of adopting a KBE approach to integrate CAD models augmented with automatic calculations to ensure product reliability are disclosed, and options for extending the work to consider more complex engineering processes are proposed.
Keywords: parametric model, KBE, knowledge Fusion, CAD, product design, customized product, hose cable, AFOSM, monte Carlo method
DOI: 10.26102/2310-6018/2024.46.3.014
The work is devoted to solving the problem of optimizing the rendering of computer three-dimensional graphics, namely the rendering pipeline. This work reduces the mentioned problem to a multidimensional version of the well-known combinatorial optimization knapsack problem. The central element of this optimization is capacity, which in the current context is the user's limited hardware capabilities, and the items to be placed in the capacity, which are various pixel shaders. The capacity is limited by the values of the hardware resources, and the shader items have two properties - utility, expressed in some value of contribution to the quality of render, and weight, which is their computational cost. The main challenge in such a context is to develop a system that will be able to solve such a knapsack problem in real time, in order to determine at each moment the best possible combination of shaders. Thus, it will be possible to obtain the best image quality and avoid downtime or overloading of the hardware. The main application of the described system will be in the sphere of computer games, web advertising, movie making and other spheres using computer graphics. Among the key problems in the development of the described system is the difficulty in determining the contribution of each individual shader to the result, due to the it’s subjectivity. Another difficulty is finding a compromise between the accuracy of the knapsack problem solution and the speed of obtaining the result, taking into account the condition that the system must work in real time and not slow down the program for which the optimization is being performed.
Keywords: knapsack problem, rendering, 3D graphics, render pipeline, optimization, neural networks
DOI: 10.26102/2310-6018/2024.46.3.015
Assessment of the effectiveness of the security monitoring and management centers is an urgent task, the solution of which depends on both the reliability of the entire system and monitoring and forecasting. The purpose of the work is to conduct a systematic analysis of factors and metrics (indicators) affecting the maturity level of monitoring centers. This problem is realized by identifying control parameters and predicting (modeling) the stability of risk management of centers when servicing requests. In particular, the formation of an integral stability index is of interest. The hypotheses of the study are considered an acceptable "tolerance band," control stability, attack planning and vulnerability analysis, the need for situational modeling. Methods of system analysis and synthesis, modeling, probability theory, heuristic approach were used. The main results of the article: 1) analysis of the sustainability of information and economic security policies and classification of direct and indirect threats in the digital business ecosystem; 2) based on the analysis done, an adaptive scheme for modeling the risk stability of a corporate system and a formal optimization model for predicting sustainable protection (based on the cost of ensuring the required security measure) were proposed; 3) as practical applications, a probabilistic model of servicing requests in a distributed system (at a given intensity of "mixing" requests of intruders) and a heuristic procedure for assessing the level of stability monitoring are proposed. The work is developed in the direction of complication of models, their elasticity and "depth" of risk accounting.
Keywords: assessment, sustainability, maturity, information security center, monitoring, risk, management
DOI: 10.26102/2310-6018/2024.46.3.013
The article is devoted to the use of artificial neural network technologies to identify objects in medical images, including images of human internal organs obtained as a result of a computed tomography procedure. The purpose of this study was to select a method for analyzing medical images and its implementation in a decision support system in surgery and urology when diagnosing human urolithiasis. The article examines the applicability of classification, detection and segmentation methods for solving various problems of object detection in medical images. It has been shown that detection is best suited for use in a medical decision support system for diagnosing urolithiasis for the purpose of planning further surgical intervention. Therefore, the article discusses the main modern neural network architectures applicable to solving the detection problem. To solve the problem of detecting objects in medical images obtained from the results of computed tomography of human internal organs, the feasibility of using a neural network of the YOLO architecture is justified. Based on the results of a computational experiment, problem areas associated with the detection of kidney objects and stones by the YOLO network were identified. To increase the accuracy of the method, it is proposed to use an algorithm for fuzzy estimation of object detection results using a neural network of the YOLO architecture. The results of image detection by the YOLO neural network after its modification allow further calculations of the parameters of the found objects for planning surgical intervention.
Keywords: computer vision, medical images, classification, detection, segmentation, neural networks, computed tomography, urolithiasis
DOI: 10.26102/2310-6018/2024.46.3.012
This article presents a new algorithm for visual data augmentation based on statistical methods. The method includes an original approach to encoding contours as one-dimensional vectors, storing information about distances from the center of gravity to vertices at specific angles. An algorithm for generating new contours is proposed, based on the statistical characteristics of the original dataset and normal distribution. The key feature of the method is the preservation of important statistical properties of the original dataset, which is confirmed by mathematical proofs of two main statements about the invariance of mathematical expectation and variance. A visual example demonstrating the method's performance on a real contour is presented. The proposed approach has potential applications in various fields, including computer vision, medical imaging, and remote sensing, where generation and augmentation of object contour data play a crucial role. The method can be particularly useful in situations where collecting real data is difficult or resource-intensive. The main results were obtained through an analytical method – the developed mathematical model is supplemented by a random number generator from a distribution with parameters calculated based on the training dataset. The parameters are selected in such a way that the main statistical characteristics of the training dataset are preserved in the synthetic data, allowing for effective application of the proposed algorithm to a wide class of pattern recognition tasks.
Keywords: contour generation, polar representation, data augmentation, computer vision, statistical characteristics, machine learning
DOI: 10.26102/2310-6018/2024.46.3.011
Managing complex logistics processes of modern enterprises requires the development of adequate mathematical models that make it possible to calculate optimal transportation plans. One of these models is the transport problem with fixed surcharges, the feature of which is the nonlinearity of the goal function. This study is devoted to the development of a genetic algorithm for solving a transport problem with fixed surcharges. The basis of the study is the analysis of existing approaches to solving various modifications of transport problems. A feature of the proposed algorithm is the formation at each stage of chromosomes that satisfy the constraints of the problem, which allows reducing the solution time. The study presents in detail the steps of the algorithm for forming the initial population, crossing over and mutation, adapted to the conditions of the transport problem with fixed surcharges. The formation of the initial population is based on the approach of randomly selecting a “supplier-consumer” pair, which ensures its sufficient diversity. The crossing over operator is implemented by developing an algorithm based on dividing modulo two the sum of the genes of the parents and subsequent redistribution of the remainders from the division between the descendants. The chromosome mutation algorithm is based on changing the transportation plan for randomly selected rows and columns while maintaining the admissibility of the individual. To conduct a computational experiment, a software product was developed in Python, and a demonstration example of the calculation is given. The results of the calculations for a group of agricultural producers allowed us to draw conclusions about the practical significance of the proposed algorithm and identified the possibilities of its use for solving multi-stage transport problems that are relevant for large manufacturing and logistics companies.
Keywords: transport problem, transport problem with fixed surcharges, genetic algorithm, chromosome, mutation, crossing over, heuristic algorithm, transportation plan, optimization
DOI: 10.26102/2310-6018/2024.46.3.020
The materials of the article are intended for specialists in the field of machine learning for the organization of technologies for improving the quality of information perception and interpretation of measurements in the practice of making medical decisions. The article proposes a method for selecting, tuning and testing a classifier under conditions of a deficit of a priori information in the data used. This is relevant when small samples of measurements of biological objects and their systems are formed at the initial stage of scientific research, the nonlinear properties of which often lead to the failure of statistical criteria. Nevertheless, the consistency of "inconvenient" distributions should be expressed in an adequate response of the program for assisting a medical decision. Based on this, the goal is determined - the choice of a solution method from the proposed set of methods for machine tuning of feature separation. Most tuning algorithms are heuristic, where the stop of parametric estimation is based on the criteria of entropy minimization as an indirect maximization of the received information. The main task is to determine the algorithm for learning and tuning the classification regression with an explicit predictive behavior of the similarity of the statistical convergence of the minimized errors. This guarantees an increase in the quality of risk classification with a trivial increase in training instances. The peculiarity of the case under consideration lies in the duality of the nature of chronic hepatitis C (CHC) progression in children: with HIV coinfection and CHC itself. This raises the problem of finding unified conditions for metric minimization of errors in еstimation the risk of developing CHC based on machine learning methods. Data sets were studied on small samples in order to identify significant parameters for heuristic identification of the presence of risks of developing the main and concomitant diseases. In this study, several methods of shallow machine learning of linear regressions were used, mainly using heuristic solutions for probabilistic separation of features. The article selectively describes the use of some basic learning methods taking into account their features in the technological verification of risk classifiers.
Keywords: machine learning, chronic hepatitis C, HIV coinfection, binary classifiers, lasso regression, sum of squared errors (MSE), regularization, decision Tree Classifier, ROC curve, area Under Curve (AUC)
DOI: 10.26102/2310-6018/2024.46.3.004
The relevance of the research conducted in the article is due to the need to predict the risks that threaten the life of a child with congenital heart disease in order to plan surgical interventions. The prognosis of the state of the cardiovascular system is based on a zero-dimensional model of blood circulation. To do this, it is proposed to create a quasi-stationary model in which the parameters of the cardiovascular system vary depending on the age of the child. Based on the analysis of data from regional monitoring of children's health, the article formulates a hypothesis that the parameters of a child's body change exponentially depending on age. Experimental studies based on monitoring data have confirmed the hypothesis put forward. A method for constructing changes in the parameters of a child's cardiovascular system depending on age is proposed and investigated. To establish this dependence, it is sufficient to have the parameter value for the child at the current time and at another time for the average child of a given gender. An algorithm for obtaining an experimental exponential dependence based on the use of Newton's iterative method for solving a nonlinear equation is proposed. The implementation of the proposed technique makes it possible to predict the state of the child's cardiovascular system for planning such interventions as surgical removal of congenital heart defects.
Keywords: congenital heart disease, approximation, exponential law, mathematical modeling, quasi-stationary system, cardiovascular system
DOI: 10.26102/2310-6018/2024.46.3.009
The article discusses the idea of increasing the efficiency of the process of servicing requests in peer-to-peer distributed computing systems based on the logical combination of their subset into peer-to-peer systems, and also proposes an algorithm for mutual information coordination of elements of the integrated system for servicing a flow of high-intensity requests for data based on the auction model. An auction model is proposed as a method and model that provides support for decentralized interaction between elements of a peer-to-peer system. The choice of the auction model – the inverse Vickrey auction model – is justified. Using the theory of multi-agent systems, an approach for the process of forming a logical group of elements of a peer-to-peer system is considered, and the corresponding software agent modules are identified that provide the functions of initializing and implementing the auction process. Using a set-theoretic representation, parameters are determined that form the conditions for the participation of nodes participating in the auction in the process of mutual information coordination, such as a cost function and a utility function. The choice and justification of the functions of the components of the auction model are considered in detail. The type of cost function and utility function used by the nodes participating in the auction is determined. Based on the composition of the functional components of the peering system elements included in the logical group, as well as determining the composition and type of functions implemented by these components, an algorithm for implementing the Vickrey reverse auction model has been developed, ensuring the formation and functioning of a logical group of peering system elements.
Keywords: distributed systems, data delivery system, peer-to-peer systems, queuing system, auction model
DOI: 10.26102/2310-6018/2024.46.3.005
The redistribution of virtualized computing and communication resources in data centers is a significant problem in the context of cloud technologies, making it difficult to ensure the stable functioning of services. These services must meet the criteria for quality of service, performance evaluation, and terms of service contracts imposed by cloud service providers. The main goal of the redistribution of virtualized computing and communication resources is the optimal placement of a subset of active virtual machines on a minimum number of physical machines, taking into account their multidimensional needs for computing and communication resources. Which will significantly improve the efficiency of a virtualized data center. The problem of redistributing computing and communication resources of a data processing center falls under the class of problems defined as "NP-hard" problems, since it involves a vast space of solutions. Therefore, more time is needed to find the optimal option. In previous studies of a number of such problems, it has been proven that metaheuristic strategies make it possible to find acceptable solutions in a suitable time. The article proposes to use a modified version of the ant colony metaheuristic algorithm to solve the problem of redistributing computing and communication resources between virtual machines of a data processing center, considered within the framework of the multidimensional vector packaging problem.
Keywords: virtualized computing and communication resources, metaheuristic methods, multidimensional vector packaging, ant colony optimization algorithm, data processing center
DOI: 10.26102/2310-6018/2024.47.4.002
The article discusses the features of using a software robot to support decision-making in an organization under uncertainty. It is noted that one of the technologies that can significantly improve the efficiency of user interaction with digital devices is smart assistants. It is proposed to consider the concept of a "smart assistant" as a software robot capable of analyzing data, identifying patterns and providing recommendations to the user on decision-making. Promising areas of application of a smart assistant are proposed depending on their functional features and possible communication channels with the user. A list of possible options for synchronizing a smart assistant with corporate information systems is considered. Models that allow a smart assistant to provide decision support functionality are considered. A comparative analysis of two options for organizing a decision support module is carried out and practical recommendations are proposed for using a combined option that involves the use of various approaches to ensure highly efficient operation of a smart assistant. A generalized internal structure of a smart assistant in a graphical format and a variant of classifying software robots according to various criteria are proposed. The proposed version of a smart assistant allows achieving high accuracy and reliability of recommendations while maintaining the speed of the software robot's response to user requests and can be used to create specialized decision support systems adapted to the specific needs of an organization.
Keywords: information system, software robot, generalized structure, classification of software robots, operational consultant, smart assistant, decision making
DOI: 10.26102/2310-6018/2024.46.3.010
The paper discusses the multilateration method to ensure coordinated interaction of unmanned aerial vehicles as a part of a swarm during monitoring of fields in agriculture, checking of the environmental parameters, collecting of the weather data etc. Multilateration will improve the reliability of the control of unmanned aerial vehicles as a part of a swarm and will ensure the autonomy of the actions of individual vehicles. The goal of the work is to assess the potential of using the radiosignal multilateration method to determine the relative position of unmanned aerial vehicles and to create of the software and physical models to test this method. In order to achieve this goal, the work presents the algorithm for the interaction of unmanned aerial vehicles using the multilateration method, a method for solving the problem of determine the location of a signal source at low computational costs and the results of computer and physical modeling of the proposed approaches. The developed models demonstrated their adequacy to the set tasks and revealed some shortcomings of the proposed approach in practical implementation. The work also examines possible situations during the interaction of unmanned aerial vehicles in a swarm and notes the main ways to eliminate shortcomings.
Keywords: multilateration method, unmanned aerial vehicle, swarm of aircraft, location determination, physical model, system of equations
DOI: 10.26102/2310-6018/2024.45.2.046
The work is devoted to the formation of principles for constructing components of a monitoring environment for managing multifunctional intelligent systems. The relevance of the topic under study is substantiated, the goal and objectives of the work are set. The task of forming a system of indicators describing the operation of the system is highlighted as a key task in the formation of a monitoring environment. Three stages are described that determine the formation of a system of indicators from system performance indicators to performance indicators of individual elements. A system of indicators for the monitoring environment is proposed in the form of a hierarchical structure with 3 levels: the level of performance criteria, the level of performance indicators, the level of a combination of types of resources and types of activities. Algorithms for collecting and generating data sets are proposed. The algorithm for generating a data set for the monitoring environment involves obtaining data from different sources. The task of the data collection algorithm is to prepare data sets for subsequent processing and obtain the values required by the monitoring environment. When collecting data, various approaches to generating target data sets may be considered. To determine the correspondence between functional areas, resources, types of activities, divisions and performers, an algorithm for generating correspondence directories is attached. The architecture of a web application is proposed as one of the forms of implementation of the monitoring environment. Using the example of using the Next.js framework, the components of the application architecture are described and an architecture diagram is presented.
Keywords: management, data, monitoring, architecture, algorithm, intelligent systems
DOI: 10.26102/2310-6018/2024.46.3.002
The presented article examines an innovative algorithm for assessing the attractiveness of potential partners in the context of online dating. The algorithm employs two neural networks: a generative network and a convolutional network. The generative neural network creates visual profiles based on various attractiveness parameters, while the convolutional neural network analyzes and extracts these parameters from images of real users. This approach allows for the dynamic adaptation of user preferences, ensuring high relevance of recommendations even with a limited pool of candidates in a given region. The method described in the article aims to significantly enhance the user experience and increase the success rate of online dating. By utilizing neural networks, the algorithm can account for individual user preferences and adapt to them in real-time. This makes the recommendations more accurate and personalized, which in turn facilitates the creation of deeper and higher-quality interpersonal connections. The research also emphasizes the importance of forming stable and happy long-term relationships. The presented approach contributes to this by providing users with a more satisfactory and effective experience in online dating. Thus, the use of algorithms and neural networks in the field of online dating has the potential to greatly improve the quality of interactions and interpersonal connections, which is a crucial aspect in the modern digital age.
Keywords: neural networks, attractiveness, online dating, generative neural network, convolutional neural network, matchmaking, recommendations, user preferences, relevance
DOI: 10.26102/2310-6018/2024.46.3.008
The relevance of the study is due to the need to systematize key skills and knowledge for effective scientific activity in research organizations. In this regard, this article aims to develop a competency model for scientific workers. The leading approach to the study of this problem is the method of integrating the competency model into the grading system, which allows for a comprehensive consideration of the assessment and stimulation of professional growth of employees. The article presents: a competency model for scientific employees of research organizations, including four main categories of competencies: professional, personal, interpersonal and managerial, detailed by skill level; a methodology for integrating a competency model into an automated grading system, providing an objective assessment and stimulation of professional growth; stages of creating a model, including identifying job groups, calculating a grading scale, differentiating skill levels and adapting the model based on feedback. A competency model proposed as a set of competencies, which, according to the heads of research organizations, are considered as indicators of behavior with the help of which scientists are able to effectively and efficiently perform their job duties. The materials of the article are of practical value for research organizations, contributing to effective talent management, career planning and development of training programs.
Keywords: competency model, grading, scientific employees, professional development, HR processes, interdisciplinary interaction, talent management
DOI: 10.26102/2310-6018/2024.45.2.020
This article discusses the problems of using neural networks of the ART family to optimize the decision-making process in risk management systems. The advantages of this approach, such as the ability to quickly respond to new information and flexibility in learning, are weighed against disadvantages, including the difficulty of adjusting parameters and interpreting results. The next part of the article will explore various ways to train ART networks, including unsupervised learning and supervised learning methods, as well as key points for configuring network parameters. Possible problems related to the quality of input data and the difficulty of interpreting output data are raised. The article also presents a concrete example of the use of ART-type neural networks in the construction industry to assess risks and make informed decisions. In conclusion, the article focuses on the prospects for using neural networks of the ART family for cluster analysis of risks, identifying related factors and grouping them for more effective management. The possibilities for further development of decision-making methods in risk management using neural networks such as ART and their potential to provide more accurate and predictive practices are discussed.
Keywords: ART-type neural networks, risks, decision-making processes, monitoring data, neural network training
DOI: 10.26102/2310-6018/2024.45.2.045
The relevance of this research stems from the fact that controlling a drone using hand gestures is more natural and intuitive than using traditional joysticks. This allows users to easily learn control and focus on task execution rather than technical aspects of operation. In turn, developing a gesture recognition system requires advancements in machine learning-based image processing algorithms. This paper aims to investigate the feasibility of implementing drone motion control using hand gestures in conjunction with modern neural network technologies. The main approach in addressing this problem involves the application of convolutional artificial neural networks for image processing and computer vision tasks. The work also explores methods for hyperparameter optimization using the Optuna tool, the use of TensorFlow Lite for implementing machine learning models on resource-constrained devices, and the application of the MediaPipe library for gesture analysis. Technologies such as Dropout and L2-regularization are used to enhance model efficiency. The materials presented in this paper hold practical value for researchers in the fields of artificial intelligence and robotics, software developers, and companies involved in the development of unmanned aerial vehicles.
Keywords: quadcopter, hand gestures, computer vision, convolutional neural networks, artificial neural networks, hyperparameter optimization, control
DOI: 10.26102/2310-6018/2024.46.3.003
The paper considers a method in which SWOT analysis is combined with the hybrid assessment method. SWOT analysis includes the identification of internal strengths and weaknesses of the organization and external opportunities and threats, which allows to choose strategies to maximize benefits and minimize risks. In turn, the hybrid assessment method combines the advantages of several well-known methods for increasing the efficiency and convenience of the decision-making process. The main idea of the method is the combined use of the hierarchy analysis method and the statistical method of calculating the weighted average, which makes it possible to combine their strengths and at the same time minimize the disadvantages. The analytical hierarchy process allows one to structure complex tasks in the form of a hierarchy, which is then formed into separate levels. Paired comparisons of hierarchy elements make it possible to assess the relative importance of each element, which provides a systematic approach to decision-making process. The purpose of the integration was to combine the positive features of the two methods. Within the framework of this article, one of the main disadvantages of the combined use of SWOT analysis and the hierarchy analysis method was identified and described in detail. A comparative analysis of the number of required pairwise comparison operations was also carried out between the combined use of SWOT analysis and the hierarchy analysis method, and the use of SWOT analysis in conjunction with the hybrid assessment method.
Keywords: decision making method, hierarchy analysis method, hybrid assessment method, non-functional requirements, functional requirements, SWOT analysis
DOI: 10.26102/2310-6018/2024.45.2.019
The relevance of this work is associated with the expanding use of information systems and models that make it possible to monitor the dynamics of key indicators of the functioning of enterprises and make appropriate organizational and managerial decisions. When working with enterprise information models, it is necessary to access data arrays, which can lead to problems with time for data analysis and query processing. When considering this task, it is important to take into account the size and structure of the basic information arrays storing the basic data of the enterprise. In this regard, this paper examines the feasibility of combining arrays that reflect the state of objects in certain workshops of a machine-building enterprise. It is shown that the gain from such an operation is possible by reducing the time of operations with the array. A problem is proposed for finding the optimal structure of the composition of the resulting base arrays, characterized by the optimal updating time. To solve this problem, an algorithm is proposed for combining the main arrays. An analysis of the feasibility of the merger process is carried out, as a result of which the conditions under which such a merger is advisable are determined. For the algorithm, it is proposed to use the “branch and bound” method. The proposed algorithm allows you to make the optimal decision on the choice of the composition of the base arrays and allows you to combine the base data arrays of the enterprise information model, ensuring a reduction in the total time of accessing the data.
Keywords: information model of an enterprise, information array, data integration, data analysis, optimization criteria, efficiency of combining information arrays, enterprise management, production organization, automation
DOI: 10.26102/2310-6018/2024.46.3.028
The article discusses the development of a mobile gaming application for the formation and development of leadership qualities in high school students, college and technical school students. The educational gaming application corresponds to the concept of "innovative educational technology", that is, it includes a set of three interrelated components: modern content, modern teaching methods, modern digital learning infrastructure. The developed mobile application allows you to systematically develop such leadership qualities as self-confidence, responsibility, time management skills, creativity, the ability to act in a situation of uncertainty, and determination. The application logic is based on the principle of forming an individual educational trajectory. To build individual learning trajectories for each user, neural network clustering of questionnaire data is used. That is, when generating individual trajectories for the development of leadership qualities, not only questionnaire methods are used, but also the result of applying clustering methods to a set of questionnaires. Self-organizing Kohonen maps are used for clustering. The resulting division into clusters was analyzed by experts, several clearly defined clusters were identified, for each of which a model of individual change in the trajectory of development of leadership qualities was compiled. As a result of expert analysis of the clustering results, seven clusters were identified. A description of each cluster was compiled jointly with experts.
Keywords: clustering, kohonen network, leadership qualities, individualization of learning trajectory, mobile application
DOI: 10.26102/2310-6018/2024.45.2.021
The article shows the possibilities of using machine learning methods to build and analyze an authentication system based on the dynamics of keystrokes. The paper substantiates the need to improve the multifactor authentication system. A method of classifying the work of behavioral biometrics for comparison and use of research results is proposed. The basic possibilities of processing and generating dynamic and static signs of the dynamics of keystrokes are considered. Various combinations of feature sets and training samples were tested, and the best combination with an Equal Error Rate (EER) of 4.7% was described. An iterative analysis of the quality of the system allows us to establish the importance of the first characters of the input sequence, as well as the nonlinear relationship between the degree of ranking of the model and EER. The high performance achieved by the boosting model indicates the significant potential of behavioral authentication for further improvement, development and application. The significance of this method, its practical usefulness not only in the task of authentication, development prospects, including the use of neural network methods and data dynamics analysis are presented. Despite the achieved results, there is a need for further work on the model, including the development of additional clustering, classification models, changing the set of features and building a cascade. The importance of the research area, which can make a significant contribution to the development of information security and technology, is emphasized.
Keywords: authentication, behavioral biometrics, keystroke dynamics, classification, machine learning
DOI: 10.26102/2310-6018/2024.45.2.013
The article presents algorithms for reconstruction, calculation of stone parameters and visualization of three-dimensional kidney and stone objects based on data obtained after the detection of 2D objects by a neural network on medical images obtained as a result of computed tomography of human internal organs. The algorithms allow you to restore (assemble) kidney and stone objects, calculate the physical parameters of stones, and perform flat and three-dimensional visualization of stones. The implementation of algorithms in the software code allows you to obtain the dimensions of the found concretions in the kidneys, visualize the density distribution inside the stone, visualize the location of the found stones in the kidney, which simplifies the support of medical decision-making during diagnosis and subsequent planning of surgical intervention during the stone crushing procedure using a laser installation. The proposed algorithms and models were implemented in a prototype of a medical decision support system in surgery and urology using computer vision technologies as part of software modules. The use of the developed algorithms for layered assembly of stones and kidneys in the prototype of a medical decision support system in surgery and urology using computer vision reduces the time for diagnosis and planning of stone crushing surgery, and also helps to avoid errors in determining the location of stones inside the kidney and, thereby, reduce the likelihood of injury to the patient.
Keywords: detection, visualization, 3D voxel reconstruction, DICOM images, YOLO network
DOI: 10.26102/2310-6018/2024.45.2.016
The article explores the possibilities of applying semantic analysis of user posts on the social network VKontakte for monitoring and predicting depression. It emphasizes the seriousness of the depression issue, its negative impact on health and society, and the relevance of early diagnosis and assistance. The study also justifies the necessity and prospects of analyzing data from Russian-language social networks to prevent the development of depression among users. The article examines the analysis of textual data and the use of logistic regression to classify users based on the presence of depression. The study's results show high model accuracy using logistic regression, demonstrating the potential for automating the processes of identifying and supporting users suffering from depression in the online environment based on user information from social networks. The significance of this method is also highlighted, along with its practical usefulness for personalized interventions, its advantages, and its development prospects, including the use of neural network methods and the analysis of data dynamics. Despite the results achieved, there is a need for further work on the model, including the study of other machine learning methods and taking into account changes in the user’s mental state over time. The development of depression prediction methods based on social network data, as proposed in the article, is an important direction that can make a significant contribution to psychology, healthcare, and information technology.
Keywords: forecasting, depression, psychological disorder, logistic regression, classification, social network, machine learning
DOI: 10.26102/2310-6018/2024.45.2.044
The work is devoted to the problem of planning ship routes in water areas with heavy traffic. In conditions of heavy traffic, navigational safety can be ensured only if ships adhere to a certain traffic pattern. The paper examines the problem of planning a route in such a way that it corresponds to the shipping practices that have developed in a particular area. The route planning method proposed in this work is based on clustering data on vessel traffic. The selected clusters represent areas in three- or four-dimensional phase space with similar speeds and courses of vessels, on the basis of which a graph of possible routes is formed. A feature of the approach for constructing a graph is the reduction in the number of vertices and edges by identifying the location of the selected clusters by covering polygons. The work shows that in many cases not only concave, but also convex polygons can be used, which can further reduce the power of the graph. The paper provides a metric for the distance between points in phase space, which is used to cluster data, and discusses the problem of choosing metric parameters and the clustering algorithm. The promise of using the DBSCAN algorithm is noted. The work is accompanied by calculations of planned vessel routes based on data from real water areas (Tsugaru Strait). The results of clustering traffic data, identifying the location of clusters by constructing enclosing polygons, and calculating the route of the vessel are presented. It is noted that the problem under consideration may be promising in the context of the future development of autonomous vessels navigation. In this case, the calculated route of the vessel will correspond to the movement of other vessels that were previously in the water area. This will reduce the likelihood of dangerous situations occurring when an autonomous vessel moves in the general traffic flow.
Keywords: navigation safety, vessel traffic control, traffic route establishment system, heavy traffic, route planning, clustering, graph algorithms
DOI: 10.26102/2310-6018/2024.45.2.043
The article considers the task of building a tourist route with predetermined points of the beginning and end of the route. The objects are divided into two types. The first ones are mandatory, which should certainly be included in the resulting route. And the second ones are additional ones, which are not necessary to visit. The route is formed taking into account the priorities set for the objects by the tourist, based on his interests and preferences, while the total time of visiting the objects should not exceed the specified deadline for arrival at the end point of the route. To solve this problem, the article proposes an approach based on the construction of a route by known methods along the main objects and its further expansion using ant strategies. To this end, the concept of "satiety" of the ant and the probability of returning to the main route are introduced, so that it is possible to control the time reserve. At the end of the article, we present the results of a computational experiment aimed at assessing the influence of the ant algorithm parameters on the resulting route and developing recommendations for adjusting these parameters depending on the size of the problem. In addition, a comparative analysis of the routes obtained by the proposed algorithm and the exact branch-and-bound method for a given set of objects is carried out, based on the results of which a conclusion is drawn about the effectiveness of the proposed algorithm.
Keywords: tourist route, ant algorithm, priority, traveling salesman's task, probabilistic choice
DOI: 10.26102/2310-6018/2024.45.2.042
The relevance of the study is due to the low level of use of dialogue in natural language in distance learning. The creation of such tools based on artificial intelligence will make the process of distance learning more accessible and attractive. The article proposes to build a dialogue based on standard questions for the content of the distance learning course. The answer is selected based on the similarity of the user's question to the standard. It is recommended to use the structural units of the distance learning course as a set of answers, and the corresponding headings as standard questions. The training dialogue data is remembered and used to expand the list of standard questions and train the system. To control learning, a measure of the similarity of the student’s answers to test questions and the correct answer options is used. To generate test questions, you can use distance learning dictionaries and test tasks. It is proposed to determine the measure of similarity of two texts using the cosine of the embeddings of the closest terms. Data from comparing texts using the proposed methodology confirm its ability to correctly assess the similarity of texts and justify its use for organizing dialogue in natural language in distance learning.
Keywords: distance learning, ranking chatbot, natural language dialogue, embedding, soft testing, sentence similarity measure
DOI: 10.26102/2310-6018/2024.45.2.041
The technology of simultaneous multithreading is considered to be of little use in programs involved in intensive calculations, in particular when multiplying matrices - one of the main operations of machine learning. The purpose of this work is to determine the limits of applicability of this type of multithreading to high performance numerical code using the example of block matrix multiplication. The paper highlights a number of characteristics of matrix multiplication code and processor architecture that affect the efficiency of using simultaneous multithreading. A method is proposed for determining the presence of structural limitations of the processor when executing more than one thread and their quantitative estimation. The influence of the used synchronization primitive and its features in relation to simultaneous multithreading are considered. The existing algorithm for dividing matrices into blocks is considered, and it is proposed to change the size of blocks and loop parameters for better utilization of the computing modules of the processor core by two threads. A model has been created to evaluate the performance of executing identical code by two threads on one physical core. A criteria has been created to determine whether computationally intensive code can be optimized using this type of multithreading. It is shown that dividing calculations between logical threads using a common L1 cache is beneficial in at least one of the common processor architectures.
Keywords: simultaneous multithreading, matrix multiplication, computation intensive, microcore, BLAS, BLIS, synchronization, cache hierarchy, spinlock
DOI: 10.26102/2310-6018/2024.45.2.015
The problem of allocation and operation of parking spaces is an important part of research in the field of intelligent transportation. In recent years, due to the sharp increase in the number of cars, the problem of limited parking space resources has been expressed. Effective parking management requires analysis of huge amounts of data and modeling to optimize the use of parking spaces. The implementation and operation of smart paid parking space in Vladivostok creates an interesting application area for data mining and machine learning. The study uses a large-scale data set of historical parking transactions in Vladivostok, including vehicle type, time, location, session duration, and more, to create a data model that reflects the relationship between parking prices, demand, and revenue. The article describes the mechanism for creating a data model that includes all important aspects of the functioning of paid parking lots and factors affecting occupancy. Using this model will allow for machine learning, application of models and evaluation of the effectiveness of their application. The study also identifies key factors influencing parking demand, such as time of day, day of week, location, etc. The data model and insights gained from this research can be used by governments and property owners to optimize the use of paid parking and improve traffic management in smart cities. The approach presented in this article can be applied to other cities to create data-driven pricing systems that meet the specific needs and characteristics of each city.
Keywords: modeling, paid parking lots, data analysis, gaussian distribution, optimization
DOI: 10.26102/2310-6018/2024.45.2.040
The article discusses choosing a technological approach to porting a Windows desktop application that utilizes a non-cross-platform user interface component library, and that implements a plugin architecture, to Linux. The approach described can be used in cases when flexibility and low overhead is preferred over a ready-made solution. The work has been done based on systems analysis. A collection of existing options and their elements is examined. The resulting solution consists in using model-driven software development to separate platform-specific components from cross-platform ones by means of well-defined programming interfaces. The suggested version of a technology by which source code is generated from a declarative description of an object-oriented interface model provides interoperability between objects, residing in different modules and separated by a compiler or a runtime library boundary. The XML technology stack is used to implement validation, code completion and transformation of model descriptions into C++ source code. Interfaces are represented by virtual method tables. Each method is a C-style function. A reference to an interface is a structure containing a pointer to a virtual method table, and a pointer to an object instance. For each interface there is a number of declarations and definition generated: a set of function declarations, a virtual method table declaration, an interface reference structure declaration, wrappers for interface references and implementation base classes in C++. The technology is successfully applied in the development of INTEGRO geographic information system.
Keywords: plug-in architecture, object-oriented programming, application binary interface, c++, INTEGRO