The article is devoted to the scientific school developed by the first author in 1995–2012 in Yaroslav-the-Wise Novgorod State University (Veliky Novgorod, Russia). The finite practical goal of the research carried out by the school can be denoted here as the revelation of the most rational variant for sense transfer in a knowledge unit defined by a set of semantically equivalent natural-language phrases. One phrase here corresponds to the simple spread natural-language sentence (according to the “Meaning–Text” theory terminology). Knowledge formed herewith about synonymy and forms of language expression of relationships between concepts of some topical area are in demand in tasks requiring the establishment of full or partial equivalence in the meaning of both complete sentences of natural language and their combinations, and individual fragments of phrases. The results are both theoretical and practical in nature. Offered methods and their software implementations can be used for decision of a wide range of tasks of recognition and analysis of semantics of complex information objects (texts and images at first), and for lossless-in-sense information compression.

The team of the speech recognition sector of the Computing Center of the Russian Academy of Sciences has participated in the development of speech technologies since their appearance in the Soviet Union in the 1960s. During this time, several generations of researchers have carried out research in this field. Accordingly, the approaches to solve problems of speech recognition have repeatedly undergone fundamental changes: the methods based on the recognition of individual sounds and parts of words using sets of rules obtained by expert means have given place to the methods of recognition and semantic interpretation of natural continuous speech based on mathematical models trained on large data sets. The review of the results begins with a description of the approach to building speech recognition systems, which was proposed by the head of the sector V.N. Trunin-Donskoy. The hardware-software approach has long been the “calling card” of the team and played a significant role in popularizing speech recognition and realizing the benefits of developing speech technologies in the Soviet Union. Solutions based on a combination of software and specialized hardware are now standard, but were new at that time. The development of the area, the complication of problems associated with the transition to discrete speech recognition with large dictionaries have resulted in the replacement of recognition methods based on the use of systems of expert rules with methods based on classical optimization algorithms, which were proposed and improved by the staff and graduate students of the team. Understanding the importance of the tasks of collecting, classifying, and annotating representative arrays of speech data was a feature of research at the Computing Center of the USSR Academy of Sciences. The work of the team members was significantly ahead of not only domestic but also modern foreign research in this area. A relevant area of applied work in the 1970s–1980s was the use of modern methods of digital speech processing, the creation of problem-oriented tools and language means that made it possible to simulate in real time the operation of components of speech recognition and signal processing systems, and in particular, interactive methods for filtering speech signals. The consistent increase in computing power and data corpus volumes has made it possible to move to the use of probabilistic speech modeling technologies, as well as the formulation and solution of natural speech recognition problems. Methods have been worked out and data corpuses and software systems have been developed for recognizing spontaneous speech, separating voices, determining gender, identifying key words in a speech stream, and classifying the subject of a speech message. Recent studies are related to the development of computationally efficient neural network methods and models for speech recognition and processing, which are intended for use in mobile devices.

This paper examines variations of argumentative semantics for many-valued logics of nonsense and uncertainty types. The proposed formal means are used to form opinions based on argumentation and analysis of their conformity with objective knowledge in an opinion poll. The logical consistency of opinion relative to such knowledge is shown to depend on argumentative semantics, which indicates the need to select tools that are adequate to the nature of the task.

The paper substantiates a method for creating training datasets for fuzzy neural networks, which can be used to promptly obtain probabilistic estimates for the causes of abnormal critical events or incidents in diagnostic systems. The rules for converting the hypotheses on potential incident causes into intervals of defect probability in a process chain at a certain stage of continuous production are considered using belief functions. We propose a procedure for converting these hypotheses into a database of fuzzy production rules automatically, which provides training an adaptive neural network based on the Takagi–Sugeno–Kang fuzzy inference system. This makes it possible to quickly calculate a relatively accurate probabilistic estimate of a malfunction in the process chain without using expensive computing resources.

The use of intelligent methods in the mechatronics and robotics areas is a hot topic due to the need for the industry to switch to digital technologies. This paper presents an approach to machine learning of an electromechanical system control loop with an actuating mechanism. The real-time control system maintains the specified drive speed using a discrete proportional-integral controller implemented in MATLAB/Simulink. The system learning task is to maximize the actuator speed directly on the running plant while the controller parameters are adjusted automatically. Training is carried out using the simplex method and a random search algorithm. The results of the experiments are presented as illustrations of the operation of both algorithms.

The paper presents a description of the developed approach and service for the intelligent analysis of source code in Python. The service reduces the time of code verification by partial automation. The FastText algorithm is used to obtain vector representations of source code texts. A pretrained neural network language model based on the transformer architecture was used to obtain a possible assignment of a natural language function. A classifier based on the gradient enhancement algorithm was used to detect repetitive headers. The developed service checks the set of changes and publishes error reports and duplicates in the format of comments of the set of changes after publishing the set of changes in the remote Git repository.

The work describes an approach to improving the quality of training of an intelligent robot control system based on error correction. The model task is to teach the mobile robot how to move in the contrast strip. It is shown that the quality of learning can be improved by removing bad hypotheses and subsequent further training. Experiments on a robot simulator showed an improvement in the learning results when using error correction.

The recognition of local anisotropies of muon fluxes using the functions of normalized variations for matrix observations of the URAGAN hodoscope is considered. Normalized instrument functions are introduced and spatiotemporal filtration is used, which become the basis of computation of the functions of normalized variations. An algorithm of recognition of local anisotropies is implemented. An experimental study of the application of the functions of normalized variations is carried out that confirms the efficiency of the developed algorithm for recognition of local anisotropies of muon fluxes in times series of matrix observations of the URAGAN hodoscope.

This is a challenge to identify tomato plant diseases using naked eyes. In implementing an automation plantation system, a plant health classification system is necessary in monitoring the health of the plants. This study proposes a system that can classify tomato plant health into five categories of healthy, early blight, late blight, bacterial spot, and yellow leaf curl virus based on their leaves using deep learning algorithms as feature extractors. Five different pretrained deep learning algorithms (i.e. Resnet-50, AlexNet, GoogleNet, VGG16, and VGG19) were studied and compared. A Raspberry Pi coupled with a camera was proposed to capture tomato plant leaf image. After that, a support vector machine (SVM) with the extracted features was trained for the plant health classification. The results indicate that SVM coupled with ResNet-50 was the best with averaged training and testing accuracies of 98.26 and 93.33%, respectively.

This work deals with the interrelated problems of assessing the closeness of a text to the most rational (reference) form of conveying its sense and the formation of a reference text collection, in relation to which the assessment itself is performed. The texts under analysis for closeness to the semantic standard are abstracts of scientific articles together with their titles. The solution is based on the comparison of values for the 5th percentile of the empirical distribution corresponding to an array of fractions for nonzero values of the term frequency (TF) for separate phrases within each abstract relative to each document under consideration for inclusion into the reference collection. A variant for numerical estimation the significance of the abstract for calculating the mentioned percentile for candidate documents with maximum precision in the case of selection of the most significant for the reference collection is offered.