Journal influence
Higher Attestation Commission (VAK) - К1 quartile
Russian Science Citation Index (RSCI)
Bookmark
Next issue
№2
Publication date:
14 June 2026
Journal articles №4 2025
11. Artificial intelligence and machine learning in lung disease diagnosis: Software overview
[№4 за 2025 год]Authors: Bariev, I.I. (i.bariev@innopolis.ru) - Innopolis University (Director);
Abstract: The paper reviews software products that utilize artificial intelligence and machine learning in X-ray image diagnostics. Routinely viewing large volumes of images greatly increases physician fatigue and potentially increases the risk of diagnostic errors. The use of automated systems reduces the workload by providing preliminary analysis and identification of pathologies, which contributes to a more accurate and timely diagnosis. Modern software based on artificial intelligence and machine learning can detect various pathologies, such as tuberculosis, pneumonia, and tumors, with an accuracy comparable to or superior to that of experienced radiologists. Automated tools significantly accelerate the analysis of large datasets, which is particularly crucial under high-workload conditions, for instance during epidemic outbreaks when diagnostic timeliness becomes critical for disease control. Under these conditions, automated systems can serve as assistants by performing preliminary analysis, while final conclusions in complex diagnostic cases remain under the purview of medical experts. These factors drive the development of diverse software products designed for lung X-ray image diagnostics. The author examines the most widely used domestic and international software products, as well as specialized Russian programs registered with the Federal Institute of Industrial Property (FIPS). The paper evaluated the disease classification accuracy of these software products. The author also provided average values for key diagnostic metrics, such as sensitivity and specificity, and identified the programs with the best disease diagnostic metrics.
Keywords: artificial intelligence, machine learning, the software, diagnostic metrics, radiographic image analysis
Visitors: 2192
11. Artificial intelligence and machine learning in lung disease diagnosis: Software overview
[№4 за 2025 год]
Visitors: 2192
The paper reviews software products that utilize artificial intelligence and machine learning in X-ray image diagnostics. Routinely viewing large volumes of images greatly increases physician fatigue and potentially increases the risk of diagnostic errors. The use of automated systems reduces the workload by providing preliminary analysis and identification of pathologies, which contributes to a more accurate and timely diagnosis. Modern software based on artificial intelligence and machine learning can detect various pathologies, such as tuberculosis, pneumonia, and tumors, with an accuracy comparable to or superior to that of experienced radiologists. Automated tools significantly accelerate the analysis of large datasets, which is particularly crucial under high-workload conditions, for instance during epidemic outbreaks when diagnostic timeliness becomes critical for disease control. Under these conditions, automated systems can serve as assistants by performing preliminary analysis, while final conclusions in complex diagnostic cases remain under the purview of medical experts. These factors drive the development of diverse software products designed for lung X-ray image diagnostics. The author examines the most widely used domestic and international software products, as well as specialized Russian programs registered with the Federal Institute of Industrial Property (FIPS). The paper evaluated the disease classification accuracy of these software products. The author also provided average values for key diagnostic metrics, such as sensitivity and specificity, and identified the programs with the best disease diagnostic metrics.
Bariev, I.I. (i.bariev@innopolis.ru) - Innopolis University (Director);
12. Building travel routes using the apparatus of cellular automata lattice gases
[№4 за 2025 год]Authors: Mendurov, S.A. (mendurow@yandex.ru) - Zhukov Air and Space Defence Academ (Adjunct); Boykova, A.V. (alexmario@mail.ru) - Zhukov Air and Space Defence Academ (Associate Professor, Professor), Ph.D;
Abstract: The paper examines the main approaches to solving the route planning problem: graph-based methods, cell decomposition techniques, potential field algorithms, optimization strategies, and intelligent algorithm methodologies. The authors identify both general and specific limitations of each existing route planning method. In various applied fields, route planning across rough terrain under dynamic conditions is becoming increasingly critical. The quality and speed of decision-making are negatively affected by the incompleteness, inaccuracy, and ambiguity of initial information. This paper aims to automate movement decision-making for transport platforms using the lattice gas cellular automata methodology. To achieve this, the authors developed a cell decomposition-based method for solving movement tasks of object of interest in areas with difficult terrain. Each cell in the lattice represents a section of terrain that can have various characteristics, such as permeability, the presence of obstacles and other parameters. Each cell can change its parameters depending on the current situation along the route. Partitioning space into manageable sections simplifies the route planning process. The lattice gas concept is fundamentally based on an analogy with molecular behavior. Each cell can interact with its neighbors, which allows us to represent the movement of objects of interest as a random process. The claimed approach ensures the construction of the optimal route that takes into account dynamic changes in the environment. The authors presented simulation results demonstrating the developed algorithm's adaptability for real-time operation in dynamically changing environments with varying terrain permeability. The proposed method can be applied to cargo delivery planning across rough terrain, emergency response operations, and unmanned systems deployment.
Keywords: route search, cellular automatic machine, models of cellular automata, optimization problem
Visitors: 1691
12. Building travel routes using the apparatus of cellular automata lattice gases
[№4 за 2025 год]
Visitors: 1691
The paper examines the main approaches to solving the route planning problem: graph-based methods, cell decomposition techniques, potential field algorithms, optimization strategies, and intelligent algorithm methodologies. The authors identify both general and specific limitations of each existing route planning method. In various applied fields, route planning across rough terrain under dynamic conditions is becoming increasingly critical. The quality and speed of decision-making are negatively affected by the incompleteness, inaccuracy, and ambiguity of initial information. This paper aims to automate movement decision-making for transport platforms using the lattice gas cellular automata methodology. To achieve this, the authors developed a cell decomposition-based method for solving movement tasks of object of interest in areas with difficult terrain. Each cell in the lattice represents a section of terrain that can have various characteristics, such as permeability, the presence of obstacles and other parameters. Each cell can change its parameters depending on the current situation along the route. Partitioning space into manageable sections simplifies the route planning process. The lattice gas concept is fundamentally based on an analogy with molecular behavior. Each cell can interact with its neighbors, which allows us to represent the movement of objects of interest as a random process. The claimed approach ensures the construction of the optimal route that takes into account dynamic changes in the environment. The authors presented simulation results demonstrating the developed algorithm's adaptability for real-time operation in dynamically changing environments with varying terrain permeability. The proposed method can be applied to cargo delivery planning across rough terrain, emergency response operations, and unmanned systems deployment.
Mendurov, S.A. (mendurow@yandex.ru) - Zhukov Air and Space Defence Academ (Adjunct); Boykova, A.V. (alexmario@mail.ru) - Zhukov Air and Space Defence Academ (Associate Professor, Professor), Ph.D;
13. Incremental markup of 19th-century handwritten archival diaries
[№4 за 2025 год]Authors: Mestetskiy, L.M. (mestlm@mail.ru) - Lomonosov Moscow State University, Higher School of Economics (Professor, Leading Researcher), Ph.D; Zykov, V.P. (valera_zykov_2003@mail.ru) - Lomonosov Moscow State University (Graduate Student), Undergraduate;
Abstract: The paper considers the problem of preparing data for machine learning based training of recognition algorithms for old handwritten texts. The research analyzes archival documents − 19th-century diaries that comprise the national cultural heritage. Modern archives preserve diaries that comprise thousands of digital page images. These images result from scanning physical handwritten documents. The special properties of these documents are the high cultural level of the authors, large volumes of manuscripts, uniform handwriting, and uniform text formatting style. Training recognition algorithms requires annotated manuscript data. This markup involves creating a precise, verbatim transcription of a text portion. A highly qualified expert assessor, a specialist in the creative heritage of the diary author, performs such verbatim transcription. However, even for a specialist, markup requires significant effort. This process constitutes the most time-consuming element in automating archival handwritten text workflows. The objective of this research is to develop a novel methodology for expert assessors. This methodology substantially decreases the labor required for handwritten diary markup. The authors propose an approach based on iterative expansion of the marked dataset with small sequential increments. Accurate manuscript markup proceeds through two sequential phases. An existing trained algorithm first produces an automatic transcription. Subsequently, a specialist edits the algorithmic output to obtain an accurate verbatim transcript. The expert's precise markup subsequently serves for the next algorithm training iteration. The proposed approach is implemented as the Podstrochnik (Interlinear) software package. The software supports the complete data markup cycle for training handwriting recognition algorithms. This cycle includes automatic transcription, subsequent editing to obtain precise markup, and model retraining. The software underwent practical testing using diaries from F.P. Litke and A.V. Sukhovo-Kobylin.
Keywords: machine learning, incremental markup, handwritten text, manuscript transcription, interlinear transcription
Visitors: 1799
13. Incremental markup of 19th-century handwritten archival diaries
[№4 за 2025 год]
Visitors: 1799
The paper considers the problem of preparing data for machine learning based training of recognition algorithms for old handwritten texts. The research analyzes archival documents − 19th-century diaries that comprise the national cultural heritage. Modern archives preserve diaries that comprise thousands of digital page images. These images result from scanning physical handwritten documents. The special properties of these documents are the high cultural level of the authors, large volumes of manuscripts, uniform handwriting, and uniform text formatting style. Training recognition algorithms requires annotated manuscript data. This markup involves creating a precise, verbatim transcription of a text portion. A highly qualified expert assessor, a specialist in the creative heritage of the diary author, performs such verbatim transcription. However, even for a specialist, markup requires significant effort. This process constitutes the most time-consuming element in automating archival handwritten text workflows. The objective of this research is to develop a novel methodology for expert assessors. This methodology substantially decreases the labor required for handwritten diary markup. The authors propose an approach based on iterative expansion of the marked dataset with small sequential increments. Accurate manuscript markup proceeds through two sequential phases. An existing trained algorithm first produces an automatic transcription. Subsequently, a specialist edits the algorithmic output to obtain an accurate verbatim transcript. The expert's precise markup subsequently serves for the next algorithm training iteration. The proposed approach is implemented as the Podstrochnik (Interlinear) software package. The software supports the complete data markup cycle for training handwriting recognition algorithms. This cycle includes automatic transcription, subsequent editing to obtain precise markup, and model retraining. The software underwent practical testing using diaries from F.P. Litke and A.V. Sukhovo-Kobylin.
Mestetskiy, L.M. (mestlm@mail.ru) - Lomonosov Moscow State University, Higher School of Economics (Professor, Leading Researcher), Ph.D; Zykov, V.P. (valera_zykov_2003@mail.ru) - Lomonosov Moscow State University (Graduate Student), Undergraduate;
[№4 за 2025 год]
Authors: Minakov, E.I. (eminakov@bk.ru) - Tula State University (Professor), Ph.D; Grachev, A.N. (ga150161@mail.ru) - Tula State University, Central Design Bureau of Apparatus Engineering (Associate Professor, Chief Engineer), Ph.D; Basherov, M.S. (basherov@icloud.com) - Tula State University, Central Design Bureau of Apparatus Engineering (Engineer), ; Sprekher D.M. (shpreher-d@yandex.ru) - Novomoskovsk Institute of D. Mendeleev University of Chemical Technology of Russian Federation, Ph.D;
Abstract: This paper addresses the automatic classification of radar surveillance objects in pulse-Doppler surveillance radars. These systems operate under constrained observation times, high interference density, and substantial feature-space overlap between objects of different physical nature. Special emphasis is placed on detecting of small-sized and low-reflectivity targets possessing small radar cross-sections, including unmanned aerial vehicles and birds. The authors propose an approach using a feedforward multilayer perceptron neural network classifier. This network trains on an expanded feature space that includes kinematic, statistical, and amplitude characteristics. To enhance the model's generalization capability, the authors employed a combined training dataset containing real trajectories, synthetically generated data using augmentation methods, and statistical modeling results. Statistical features were calculated within a sliding time window. Testing demonstrated a classification accuracy of 96%. Empirical evaluation confirmed the model's robustness to noise and trajectory data variability. The authors conducted a comprehensive feature significance analysis. During network design, an a priori assessment of feature informativeness. After training, they implemented a posteriori analysis of feature contributions using a modified Garson's algorithm. The obtained results justify a minimal and physically interpretable feature set. The obtained results justify maximum model selectivity within acceptable computational complexity. All stages of model construction, feature space formation, and training were implemented in the MATLAB environment using the Neural Network Toolbox, Deep Learning Toolbox, and Signal Processing Toolbox. This approach ensured reproducibility, adaptability, and technical applicability of the developed solution. The practical significance of this work lies in the ability to integrate the developed algorithm into existing radar systems without major hardware modifications, making this approach relevant for adaptive recognition tasks in real-time operational environments.
Keywords: a neural network, classification, pulse-Doppler radar, artificial intelligence, automatic recognition, machine learning, program realization, matlab
Visitors: 2541
Authors: Minakov, E.I. (eminakov@bk.ru) - Tula State University (Professor), Ph.D; Grachev, A.N. (ga150161@mail.ru) - Tula State University, Central Design Bureau of Apparatus Engineering (Associate Professor, Chief Engineer), Ph.D; Basherov, M.S. (basherov@icloud.com) - Tula State University, Central Design Bureau of Apparatus Engineering (Engineer), ; Sprekher D.M. (shpreher-d@yandex.ru) - Novomoskovsk Institute of D. Mendeleev University of Chemical Technology of Russian Federation, Ph.D;
Abstract: This paper addresses the automatic classification of radar surveillance objects in pulse-Doppler surveillance radars. These systems operate under constrained observation times, high interference density, and substantial feature-space overlap between objects of different physical nature. Special emphasis is placed on detecting of small-sized and low-reflectivity targets possessing small radar cross-sections, including unmanned aerial vehicles and birds. The authors propose an approach using a feedforward multilayer perceptron neural network classifier. This network trains on an expanded feature space that includes kinematic, statistical, and amplitude characteristics. To enhance the model's generalization capability, the authors employed a combined training dataset containing real trajectories, synthetically generated data using augmentation methods, and statistical modeling results. Statistical features were calculated within a sliding time window. Testing demonstrated a classification accuracy of 96%. Empirical evaluation confirmed the model's robustness to noise and trajectory data variability. The authors conducted a comprehensive feature significance analysis. During network design, an a priori assessment of feature informativeness. After training, they implemented a posteriori analysis of feature contributions using a modified Garson's algorithm. The obtained results justify a minimal and physically interpretable feature set. The obtained results justify maximum model selectivity within acceptable computational complexity. All stages of model construction, feature space formation, and training were implemented in the MATLAB environment using the Neural Network Toolbox, Deep Learning Toolbox, and Signal Processing Toolbox. This approach ensured reproducibility, adaptability, and technical applicability of the developed solution. The practical significance of this work lies in the ability to integrate the developed algorithm into existing radar systems without major hardware modifications, making this approach relevant for adaptive recognition tasks in real-time operational environments.
Keywords: a neural network, classification, pulse-Doppler radar, artificial intelligence, automatic recognition, machine learning, program realization, matlab
Visitors: 2541
[№4 за 2025 год]
Visitors: 2541
This paper addresses the automatic classification of radar surveillance objects in pulse-Doppler surveillance radars. These systems operate under constrained observation times, high interference density, and substantial feature-space overlap between objects of different physical nature. Special emphasis is placed on detecting of small-sized and low-reflectivity targets possessing small radar cross-sections, including unmanned aerial vehicles and birds. The authors propose an approach using a feedforward multilayer perceptron neural network classifier. This network trains on an expanded feature space that includes kinematic, statistical, and amplitude characteristics. To enhance the model's generalization capability, the authors employed a combined training dataset containing real trajectories, synthetically generated data using augmentation methods, and statistical modeling results. Statistical features were calculated within a sliding time window. Testing demonstrated a classification accuracy of 96%. Empirical evaluation confirmed the model's robustness to noise and trajectory data variability. The authors conducted a comprehensive feature significance analysis. During network design, an a priori assessment of feature informativeness. After training, they implemented a posteriori analysis of feature contributions using a modified Garson's algorithm. The obtained results justify a minimal and physically interpretable feature set. The obtained results justify maximum model selectivity within acceptable computational complexity. All stages of model construction, feature space formation, and training were implemented in the MATLAB environment using the Neural Network Toolbox, Deep Learning Toolbox, and Signal Processing Toolbox. This approach ensured reproducibility, adaptability, and technical applicability of the developed solution. The practical significance of this work lies in the ability to integrate the developed algorithm into existing radar systems without major hardware modifications, making this approach relevant for adaptive recognition tasks in real-time operational environments.
Minakov, E.I. (eminakov@bk.ru) - Tula State University (Professor), Ph.D; Grachev, A.N. (ga150161@mail.ru) - Tula State University, Central Design Bureau of Apparatus Engineering (Associate Professor, Chief Engineer), Ph.D; Basherov, M.S. (basherov@icloud.com) - Tula State University, Central Design Bureau of Apparatus Engineering (Engineer), ; Sprekher D.M. (shpreher-d@yandex.ru) - Novomoskovsk Institute of D. Mendeleev University of Chemical Technology of Russian Federation, Ph.D;
[№4 за 2025 год]
Authors: Leonov, A.G. (dr.l@math.msu.su) - M.V. Lomonosov Moscow State University, Federal State Institution SRISA RAS, Moscow State Pedagogical University, State University of Management (Associate Professor, Professor, Head of Chair, Leading Researcher), Ph.D; Mashchenko, K.A. (kirill010399@vip.niisi.ru) - M.V. Lomonosov Moscow State University, Federal State Institution SRISA RAS, State University of Management (Junior Researcher); Martynov, N.S. (nikolai.martynov@math.msu.ru) - Federal State Institution SRISA RAS (Engineer); Pchelin, K.P. (konstantin.pchelin@math.msu.ru) - National Research Centre “Kurchatov Institute” – SRISA (Technical Fellow); Shlyakhov, A.V. (shlyakhov@vip.niisi.ru) - M.V. Lomonosov Moscow State University (Junior Researcher);
Abstract: The paper presents a system for a personal digital twin of a university instructor, designed for the automated creation and updating of educational video materials within a digital learning platform. The proposed system architecture consists of a set of interconnected modules, each addressing a specific task and ensuring both flexibility and scalability. The first stage implements automatic speech recognition using the open-source multilingual Whisper model. The system handles original audio recordings without requiring prior domain-specific fine-tuning. The system then processes the retrieved text through a universal preprocessor, which performs three key functions: numeral normalization, word stress prediction, and transcription into the International Phonetic Alphabet (IPA). Speech synthesis is implemented using a modified CoquiTTS architecture supporting zero-shot voice cloning from a short instructor speech sample. The LivePortrait technology generates the visual component by creating realistic video portraits of the instructor using submitted photographs. The LipSync module synchronizes lip movements with generated audio, significantly improving perceived video quality. The final stage involves the automatic assembly of a presentation that integrates the generated video segments. The paper presents a comparative analysis of alternative architectures for each module, identifies evaluation criteria for speech synthesis and video generation quality, and outlines the specifics of developing a Russian-language text preprocessor. Experimental results demonstrate that the proposed system reduces instructor workload for the creation and updating of educational materials by more than 70% while maintaining a high level of naturalness. The authors discussed limitations in text processing of formulas and abbreviations, while also examining potential optimizations for improving the performance of the LipSync module.
Keywords: digital twin, text preprocessor, educational video materials, automatic speech recognition (ASR), text-to-speech (TTS), LipSync, zero-shot voice cloning, Russian-language TTS
Visitors: 2027
Authors: Leonov, A.G. (dr.l@math.msu.su) - M.V. Lomonosov Moscow State University, Federal State Institution SRISA RAS, Moscow State Pedagogical University, State University of Management (Associate Professor, Professor, Head of Chair, Leading Researcher), Ph.D; Mashchenko, K.A. (kirill010399@vip.niisi.ru) - M.V. Lomonosov Moscow State University, Federal State Institution SRISA RAS, State University of Management (Junior Researcher); Martynov, N.S. (nikolai.martynov@math.msu.ru) - Federal State Institution SRISA RAS (Engineer); Pchelin, K.P. (konstantin.pchelin@math.msu.ru) - National Research Centre “Kurchatov Institute” – SRISA (Technical Fellow); Shlyakhov, A.V. (shlyakhov@vip.niisi.ru) - M.V. Lomonosov Moscow State University (Junior Researcher);
Abstract: The paper presents a system for a personal digital twin of a university instructor, designed for the automated creation and updating of educational video materials within a digital learning platform. The proposed system architecture consists of a set of interconnected modules, each addressing a specific task and ensuring both flexibility and scalability. The first stage implements automatic speech recognition using the open-source multilingual Whisper model. The system handles original audio recordings without requiring prior domain-specific fine-tuning. The system then processes the retrieved text through a universal preprocessor, which performs three key functions: numeral normalization, word stress prediction, and transcription into the International Phonetic Alphabet (IPA). Speech synthesis is implemented using a modified CoquiTTS architecture supporting zero-shot voice cloning from a short instructor speech sample. The LivePortrait technology generates the visual component by creating realistic video portraits of the instructor using submitted photographs. The LipSync module synchronizes lip movements with generated audio, significantly improving perceived video quality. The final stage involves the automatic assembly of a presentation that integrates the generated video segments. The paper presents a comparative analysis of alternative architectures for each module, identifies evaluation criteria for speech synthesis and video generation quality, and outlines the specifics of developing a Russian-language text preprocessor. Experimental results demonstrate that the proposed system reduces instructor workload for the creation and updating of educational materials by more than 70% while maintaining a high level of naturalness. The authors discussed limitations in text processing of formulas and abbreviations, while also examining potential optimizations for improving the performance of the LipSync module.
Keywords: digital twin, text preprocessor, educational video materials, automatic speech recognition (ASR), text-to-speech (TTS), LipSync, zero-shot voice cloning, Russian-language TTS
Visitors: 2027
[№4 за 2025 год]
Visitors: 2027
The paper presents a system for a personal digital twin of a university instructor, designed for the automated creation and updating of educational video materials within a digital learning platform. The proposed system architecture consists of a set of interconnected modules, each addressing a specific task and ensuring both flexibility and scalability. The first stage implements automatic speech recognition using the open-source multilingual Whisper model. The system handles original audio recordings without requiring prior domain-specific fine-tuning. The system then processes the retrieved text through a universal preprocessor, which performs three key functions: numeral normalization, word stress prediction, and transcription into the International Phonetic Alphabet (IPA). Speech synthesis is implemented using a modified CoquiTTS architecture supporting zero-shot voice cloning from a short instructor speech sample. The LivePortrait technology generates the visual component by creating realistic video portraits of the instructor using submitted photographs. The LipSync module synchronizes lip movements with generated audio, significantly improving perceived video quality. The final stage involves the automatic assembly of a presentation that integrates the generated video segments. The paper presents a comparative analysis of alternative architectures for each module, identifies evaluation criteria for speech synthesis and video generation quality, and outlines the specifics of developing a Russian-language text preprocessor. Experimental results demonstrate that the proposed system reduces instructor workload for the creation and updating of educational materials by more than 70% while maintaining a high level of naturalness. The authors discussed limitations in text processing of formulas and abbreviations, while also examining potential optimizations for improving the performance of the LipSync module.
Leonov, A.G. (dr.l@math.msu.su) - M.V. Lomonosov Moscow State University, Federal State Institution SRISA RAS, Moscow State Pedagogical University, State University of Management (Associate Professor, Professor, Head of Chair, Leading Researcher), Ph.D; Mashchenko, K.A. (kirill010399@vip.niisi.ru) - M.V. Lomonosov Moscow State University, Federal State Institution SRISA RAS, State University of Management (Junior Researcher); Martynov, N.S. (nikolai.martynov@math.msu.ru) - Federal State Institution SRISA RAS (Engineer); Pchelin, K.P. (konstantin.pchelin@math.msu.ru) - National Research Centre “Kurchatov Institute” – SRISA (Technical Fellow); Shlyakhov, A.V. (shlyakhov@vip.niisi.ru) - M.V. Lomonosov Moscow State University (Junior Researcher);
16. Software-algorithmic support for intelligent control systems based on soft computing technologies
[№4 за 2025 год]Authors: A.G. Reshetnikov (reshetnikovag@pochta.ru) - Dubna State University, Institute of the System Analysis and Control (Associate Professor); Ulyanov, S.V. (ulyanovsv46_46@mail.ru) - Dubna State University – Institute of System Analysis and Control, Dubna, Joint Institute for Nuclear Research – Laboratory of Information Technology (Professor), Ph.D;
Abstract: The paper presents the first stage of designing embedded self-organizing intelligent control systems using the QSCIT software toolkit. The toolkit utilizes SCOptKBTM soft computing technology and serves for designing robust knowledge bases in self-organizing intelligent control systems. The object of intellectualization is nitrogen pressure control in a cryogenic unit of a superconducting magnet test facility at a magnet factory. This magnet represents a control object possessing weakly formalized structure and numerous hidden parameters within the technological cooling process model. The foundation of intelligent control lies in the information-thermodynamic law of optimal distribution of quality characteristics such as stability, controllability, and robustness. The authors describe an operational system and its control process technology. Knowledge base formation occurs through extraction of information from the training signal. Genetic algorithm application enables obtaining this signal without using a mathematical model of the control object. The authors examine in detail the technology for designing embedded fuzzy controllers based on physically measurable training signals. The paper outlines the procedure for acquiring training signals by applying genetic algorithms to real-world experimental setups. The developed software enables configuration and control of experimental apparatus through TANGO Control system. The paper provides a comparison between fuzzy controllers developed using the “Knowledge base optimizer” software toolkit with a classical PID controller. This work analyzes both advantages and limitations of applying soft computing methodologies.
Keywords: intelligent control, fuzzy controller, generic algorithm, neural network, knowledge base optimizer, experimental equipment,, physical facility
Visitors: 2342
16. Software-algorithmic support for intelligent control systems based on soft computing technologies
[№4 за 2025 год]
Visitors: 2342
The paper presents the first stage of designing embedded self-organizing intelligent control systems using the QSCIT software toolkit. The toolkit utilizes SCOptKBTM soft computing technology and serves for designing robust knowledge bases in self-organizing intelligent control systems. The object of intellectualization is nitrogen pressure control in a cryogenic unit of a superconducting magnet test facility at a magnet factory. This magnet represents a control object possessing weakly formalized structure and numerous hidden parameters within the technological cooling process model. The foundation of intelligent control lies in the information-thermodynamic law of optimal distribution of quality characteristics such as stability, controllability, and robustness. The authors describe an operational system and its control process technology. Knowledge base formation occurs through extraction of information from the training signal. Genetic algorithm application enables obtaining this signal without using a mathematical model of the control object. The authors examine in detail the technology for designing embedded fuzzy controllers based on physically measurable training signals. The paper outlines the procedure for acquiring training signals by applying genetic algorithms to real-world experimental setups. The developed software enables configuration and control of experimental apparatus through TANGO Control system. The paper provides a comparison between fuzzy controllers developed using the “Knowledge base optimizer” software toolkit with a classical PID controller. This work analyzes both advantages and limitations of applying soft computing methodologies.
A.G. Reshetnikov (reshetnikovag@pochta.ru) - Dubna State University, Institute of the System Analysis and Control (Associate Professor); Ulyanov, S.V. (ulyanovsv46_46@mail.ru) - Dubna State University – Institute of System Analysis and Control, Dubna, Joint Institute for Nuclear Research – Laboratory of Information Technology (Professor), Ph.D;
[№4 за 2025 год]
Authors: S.A. Vlasova (svlasova@jscc.ru) - Joint Supercomputer Center of the Russian Academy of Sciences – JSCC (Leader Researcher), Ph.D; N.E. Kalenov (nkalenov@jscc.ru) - Joint Supercomputer Center of the Russian Academy of Sciences – JSCC (Professor, Chief Researcher), Ph.D; A.N. Sotnikov (asotnikov@iscc.ru) - Joint Supercomputer Center of RAS (Professor), Ph.D;
Abstract: The paper addresses research issues related to the formation of the common digital space of scientific knowledge (CDSSK). It constitutes a publicly accessible digital environment that consolidates information about various science-related objects. Such objects may include digital copies of physical entities (book texts, archival documents, museum artifacts), databases, information about researchers, scientific events, established facts, and similar items. The CDSSK architecture implements a linked data framework based on Semantic Web standards and ontological methodologies. This work addresses content for-mation for the subspaces (SS) comprising the CDSSK. The authors present a general flowchart for the data input algorithm into the SS, based on the developed ontology – which comprises attribute directories, object relationships, and static dictionaries of attribute values and relationships – and describe a dialog-based software package that implements these algorithms. The content generation algorithms for the thematic SS utilize analysis of objects from the auxiliary class Formats, which belongs to the universal SS. They serve as the basis for constructing a data entry dialogue scenario and implementing automatic formal and logical control of the information entered. The authors created the software package using Microsoft ASP.NET technology, operating on the Microsoft.NET Framework platform and coded in C#. It comprises four modules: object dictionaries generation; relationship establishment between objects and their attributes; dictionaries editing; and object viewing. For each module, the authors described a dialog scenario and provided implementation examples illustrated with screenshots.
Keywords: software package, digital space of scientific knowledge, ontology, content generation algorithms, data editing, object attributes, linked data
Visitors: 1980
Authors: S.A. Vlasova (svlasova@jscc.ru) - Joint Supercomputer Center of the Russian Academy of Sciences – JSCC (Leader Researcher), Ph.D; N.E. Kalenov (nkalenov@jscc.ru) - Joint Supercomputer Center of the Russian Academy of Sciences – JSCC (Professor, Chief Researcher), Ph.D; A.N. Sotnikov (asotnikov@iscc.ru) - Joint Supercomputer Center of RAS (Professor), Ph.D;
Abstract: The paper addresses research issues related to the formation of the common digital space of scientific knowledge (CDSSK). It constitutes a publicly accessible digital environment that consolidates information about various science-related objects. Such objects may include digital copies of physical entities (book texts, archival documents, museum artifacts), databases, information about researchers, scientific events, established facts, and similar items. The CDSSK architecture implements a linked data framework based on Semantic Web standards and ontological methodologies. This work addresses content for-mation for the subspaces (SS) comprising the CDSSK. The authors present a general flowchart for the data input algorithm into the SS, based on the developed ontology – which comprises attribute directories, object relationships, and static dictionaries of attribute values and relationships – and describe a dialog-based software package that implements these algorithms. The content generation algorithms for the thematic SS utilize analysis of objects from the auxiliary class Formats, which belongs to the universal SS. They serve as the basis for constructing a data entry dialogue scenario and implementing automatic formal and logical control of the information entered. The authors created the software package using Microsoft ASP.NET technology, operating on the Microsoft.NET Framework platform and coded in C#. It comprises four modules: object dictionaries generation; relationship establishment between objects and their attributes; dictionaries editing; and object viewing. For each module, the authors described a dialog scenario and provided implementation examples illustrated with screenshots.
Keywords: software package, digital space of scientific knowledge, ontology, content generation algorithms, data editing, object attributes, linked data
Visitors: 1980
[№4 за 2025 год]
Visitors: 1980
The paper addresses research issues related to the formation of the common digital space of scientific knowledge (CDSSK). It constitutes a publicly accessible digital environment that consolidates information about various science-related objects. Such objects may include digital copies of physical entities (book texts, archival documents, museum artifacts), databases, information about researchers, scientific events, established facts, and similar items. The CDSSK architecture implements a linked data framework based on Semantic Web standards and ontological methodologies. This work addresses content for-mation for the subspaces (SS) comprising the CDSSK. The authors present a general flowchart for the data input algorithm into the SS, based on the developed ontology – which comprises attribute directories, object relationships, and static dictionaries of attribute values and relationships – and describe a dialog-based software package that implements these algorithms. The content generation algorithms for the thematic SS utilize analysis of objects from the auxiliary class Formats, which belongs to the universal SS. They serve as the basis for constructing a data entry dialogue scenario and implementing automatic formal and logical control of the information entered. The authors created the software package using Microsoft ASP.NET technology, operating on the Microsoft.NET Framework platform and coded in C#. It comprises four modules: object dictionaries generation; relationship establishment between objects and their attributes; dictionaries editing; and object viewing. For each module, the authors described a dialog scenario and provided implementation examples illustrated with screenshots.
S.A. Vlasova (svlasova@jscc.ru) - Joint Supercomputer Center of the Russian Academy of Sciences – JSCC (Leader Researcher), Ph.D; N.E. Kalenov (nkalenov@jscc.ru) - Joint Supercomputer Center of the Russian Academy of Sciences – JSCC (Professor, Chief Researcher), Ph.D; A.N. Sotnikov (asotnikov@iscc.ru) - Joint Supercomputer Center of RAS (Professor), Ph.D;
18. Neural network-based pronunciation diagnosis for english language. Web service development
[№4 за 2025 год]Authors: Dorokhin, M.A. (michaelsagittarius08@gmail.com) - Saint Petersburg State University of Aerospace Instrumentation (Bachelor of Sci.); S.A. Chernyshev (chernyshev.s.a@bk.ru) - General of the Army A.V. Khrulyov Military Academy of Logistics, St. Petersburg State University of Aerospace Instrumentation (Senior Researcher, Senior Lecturer), Ph.D;
Abstract: This paper presents the development of a web service designed to improve English pronunciation through neural network-based phoneme recognition. The subject of the research is the automatic assessment of a learner's phonemic accuracy and provision of visual feedback about detected pronunciation errors. The solution employs deep learning methods: a pretrained Wav2Vec 2.0-based ASR module converts audio signals into phoneme sequences, while a subsequent classifier compares the resulting transcription with a reference (generated using a language phonemic model). The system identifies pronunciation errors through color-coded highlighting and initiates user repetition exercises for correction. As a research method, the authors employ experimental approaches to collect a specialized dataset containing recordings with pronunciation errors at the individual phoneme level. To evaluate model quality, the authors use standard accuracy metrics and Levenshtein distance. The obtained results demonstrate that the proposed system achieves over 90% accuracy in recognizing correct pho-nemes. The practical significance of this work lies in the potential integration of the service into online language-learning platforms and mobile applications, as well as its ability to provide personalized feedback to help learners improve their speech quality.
Keywords: intellectual system, machine learning, automatic speech recognition (ASR), MDD, CAPT, neural network, Wav2Vec 2.0, phonemic annotation, levenshtein distance
Visitors: 2201
18. Neural network-based pronunciation diagnosis for english language. Web service development
[№4 за 2025 год]
Visitors: 2201
This paper presents the development of a web service designed to improve English pronunciation through neural network-based phoneme recognition. The subject of the research is the automatic assessment of a learner's phonemic accuracy and provision of visual feedback about detected pronunciation errors. The solution employs deep learning methods: a pretrained Wav2Vec 2.0-based ASR module converts audio signals into phoneme sequences, while a subsequent classifier compares the resulting transcription with a reference (generated using a language phonemic model). The system identifies pronunciation errors through color-coded highlighting and initiates user repetition exercises for correction. As a research method, the authors employ experimental approaches to collect a specialized dataset containing recordings with pronunciation errors at the individual phoneme level. To evaluate model quality, the authors use standard accuracy metrics and Levenshtein distance. The obtained results demonstrate that the proposed system achieves over 90% accuracy in recognizing correct pho-nemes. The practical significance of this work lies in the potential integration of the service into online language-learning platforms and mobile applications, as well as its ability to provide personalized feedback to help learners improve their speech quality.
Dorokhin, M.A. (michaelsagittarius08@gmail.com) - Saint Petersburg State University of Aerospace Instrumentation (Bachelor of Sci.); S.A. Chernyshev (chernyshev.s.a@bk.ru) - General of the Army A.V. Khrulyov Military Academy of Logistics, St. Petersburg State University of Aerospace Instrumentation (Senior Researcher, Senior Lecturer), Ph.D;
