The use of machine learning when working with text documents significantly increases the efficiency of work and expands the range of tasks to be solved. The paper provides an analysis of the main methods of presenting data in a digital format and machine learning algorithms, and a conclusion is made about the optimal solution for generative and discriminative tasks.

Keywords: machine learning, natural language processing, transformer architecture models, gradient boosting, large language models

This article presents a comprehensive analysis of Russian-language texts utilizing neural network models based on the Bidirectional Encoder Representations from Transformers (BERT) architecture. The study employs specialized models for the Russian language: RuBERT-tiny, RuBERT-tiny2, and RuBERT-base-cased. The proposed methodology encompasses morphological, syntactic, and semantic levels of analysis, integrating lemmatization, part-of-speech tagging, morphological feature identification, syntactic dependency parsing, semantic role labeling, and relation extraction. The application of BERT-family models achieves accuracy rates exceeding 98% for lemmatization, 97% for part-of-speech tagging and morphological feature identification, 96% for syntactic parsing, and 94% for semantic analysis. The method is suitable for tasks requiring deep text comprehension and can be optimized for processing large corpora.

Keywords: BERT, Russian-language texts, morphological analysis, syntactic analysis, semantic analysis, lemmatization, RuBERT, natural language processing, NLP

The article presents an analysis of modern methods of image generation: variational autoencoders (VAE), generative adversarial networks (GAN) and diffusion models. The main attention is paid to a comparative analysis of their performance, generation quality and computational requirements. The Frechet Inception Distance (FID) metric is used to assess the image quality. Diffusion models showed the best results (FID 20.8), outperforming VAE (FID 59.75) and GAN (FID 38.9), but require significant resources. VAEs are stable, but generate blurry images. GANs provide high quality, but suffer from training instability and mode collapse. Diffusion models, due to step-by-step noise decoding, combine detail and structure, which makes them the most promising. Also considered are methods of image-to-image generation used for image modification. The results of the study are useful for specialists in the field of machine learning and computer vision, contributing to the improvement of algorithms and expansion of the areas of application of generative models.

Keywords: deepfake, deep learning, artificial intelligence, GAN, VAE, diffusion model

This paper examines methods for modeling the spread of infectious diseases. It discusses the features of the generalized compartmental approach to epidemic modeling, which divides the population into non-overlapping groups of individuals. The forecast of models built using this approach involves estimating the size of these groups over time. The paper proposes a method for estimating model parameters based on statistical data. It also introduces a method for estimating confidence intervals for the model forecast, based on a series of stochastic model runs. A computational experiment demonstrates the effectiveness of the proposed methods using data on the spread of influenza in European countries. The results show the model's efficiency in predicting the dynamics of the epidemic and estimating confidence intervals for the forecast. The paper also justifies the applicability of the described methods to modeling chronic diseases.

Keywords: epidemic modeling, computer modeling, compartmental models, SIR, stochastic modeling, parameter estimation, confidence interval, forecast, influenza

A non-stationary system of automatic speed control of a DC motor with an adaptive controller is considered. Comparative simulation modeling in Simulink of the system with and without an adapter is performed. The results of the modeling confirm the stability of the adaptive system in a larger range of change of the non-stationary parameter compared to the conventional system. At the same time, the speed and quality of transient processes are maintained at the level recommended for such systems.

Keywords: automatic control system, non-stationarity, adaptive controller, subordinate control system, electromechanical object, DC motor

Many problems related to high-speed interaction with soil represent an interesting area of ​​research. For example, the fall of heavy objects on the ground surface not only creates a dynamic impact effect, but can also serve as an effective method of soil compaction under future foundations of buildings and structures. This process, along with the penetration of objects into the soil, poses new challenges for researchers. The most accurate results in these complex scenarios can be obtained by using a nonlinear dynamic formulation, which allows for a deeper understanding of the interaction mechanisms and ensures the reliability of structures under extreme loads. This requires using appropriate modeling approaches. In addition, under such an impact, the soil exhibits the properties of a liquid or gas, so it is necessary to use special soil models. The paper presents the main basic relationships and main parameters of soil models required for dynamic calculations of soils, which can be useful in modeling the operation of a soil massif in modern software packages.

Keywords: physical nonlinearity, damping, soil, foundation of buildings and structures, dilatancy, soil compaction, pore pressure, soil density, deformation modulus, numerical soil model

The method of synthesis of control of a territorially distributed complex technical system with metrological support is presented. The synthesis method is based on the method for identifying the parameters of a stationary semi-Markov model of operation of a complex technical system, developed by the author, based on solving a system of algebraic equations, which includes the linear invariants of the semi-Markov stationary model identified in the article. The results of modeling changes in the parameters of a complex technical system are presented, taking into account the current state of the fleet of complex technical systems with an optimal choice of the interval between checks, rational use of redundancy and stationary maintenance. The obtained results can find application in the decision support system for managing a fleet of complex technical systems. by choosing the optimal interval between checks, using redundancy and carrying out stationary maintenance.

Keywords: park of complex technical systems, control synthesis method, system invariants

The article discusses the process of developing and modeling an impeller for an unmanned aircraft of the airplane type. Aerodynamic and strength calculations were carried out, key design parameters were determined, including the number of blades, engine power and choice of material. The developed models were created in the CAD system Compass 3D and manufactured by 3D printing using PETG plastic. Impeller thrust tests were carried out depending on engine speed, which allowed the design to be optimized for maximum efficiency.

Keywords: impeller, unmanned aircraft, aerodynamics, 3D modeling, 3D compass, additive technologies, thrust, testing, APM FEM

Predictive analytics is one of the most important areas of data analysis, which allows predicting future events based on historical data. The relevance of predictive analytics in the modern world is due to the rapid development of technology, the growth of data volumes and the growing need for informed management decision-making. The article discusses the main approaches such as regression models, time series, decision trees, clustering methods and neural networks, as well as their advantages and disadvantages.

Keywords: predictive analytics, regression models, time series, decision trees, neural networks, clustering, big data, predictive analytics methods, big data analysis, forecasting

In this paper, an inverse wavelet image transform method for JPEG XS format is proposed. The said format uses Le Gall wavelet filter and the lifting scheme is used as a wavelet transform. This method of wavelet processing of images and video signal has low computation speed. To improve the computation speed, it is proposed to use the Winograd method as this method allows parallel processing of groups of pixels. The paper analyses the impact of accuracy in obtaining high quality image for fixed point format computation. The simulation results show that processing of 2 pixels using Winograd's method is sufficient to use 3 decimal places to obtain high quality image. When processing 3 and 4 pixels of the image it is sufficient to use 7 decimal places each. When processing 5 pixels of the image it is enough to use 12 decimal places. A promising direction for further research is the development of hardware accelerators for performing the inverse discrete wavelet transform by the Winograd method.

Keywords: inverse discrete wavelet transform, Le Gall filter, Winograd method, image processing, digital filtering, JPEG XS

The article presents the results of the development of an algorithm and a desktop application for recognizing Russian-language handwritten text in images using computer vision and deep learning technologies. Classical and modern recognition methods have been studied and an algorithm has been developed and implemented that ensures 71% recognition accuracy. The application allows the user to upload images receive digitized text and save the results in his personal account. The software implementation includes a training block for the model with an assessment of accuracy and completeness metrics. The application meets all the set requirements providing ease of use and functionality.

Keywords: deep learning, handwritten text, image, data, model training, computer vision, feature extraction, CTC, RNN, CNN, CRNN

In this article, an analysis of the main methods for solving the linear cutting problem (LCP) with the criterion of minimizing the number of knife rearrangements is presented. The linear cutting problem in its general form represents an optimization problem that involves placing given types of material (rolls) in such a way as to minimize waste and/or maximize the use of raw materials, taking into account constraints on the number of knives, the width of the master roll, and the required orders. This article discusses a specific case of the problem with an additional condition for minimizing knives' changes and the following approaches for its solution: the exhaustive search method, which ensures finding a global optimal solution but can be extremely inefficient for problems with a large number of orders, as well as random search based on genetic and evolutionary algorithms that model natural selection processes to find good solutions. Pseudocode is provided for various methods of solving the LCP. A comparison is made in terms of algorithmic complexity, controllability of execution time, and accuracy. The random search based on genetic and evolutionary algorithms proved to be more suited for solving the LCP with the minimization of waste and knife rearrangements.

Keywords: paper production planning, linear cutting, exhaustive search, genetic algorithm, waste minimization, knife permutation minimization

Deviation of forestry equipment from the designated route leads to environmental, legal, and economic issues, such as soil damage, tree destruction, and fines. Autonomous route correction systems are essential to address these problems. The aim of this study is to develop a system for deviation detection and trajectory calculation to return to the designated route. The system determines the current position of the equipment using global positioning sensors and an inertial measurement unit. The Kalman filter ensures positioning accuracy, while the A* algorithm and trajectory smoothing methods are used to compute efficient routes considering obstacles and turning radii. The proposed solution effectively detects deviations and calculates a trajectory for returning to the route.

Keywords: deviation detection, route correction, mobile application, Kalman filter, logging operations

The article considers the issues of imitation modeling of fibrous material mixing processes using Markov processes. The correct combination and redistribution of components in a two-component mixture significantly affects their physical properties, and the developed model makes it possible to optimize this process. The authors propose an algorithm for modeling transitions between mixture states based on Markov processes.

Keywords: modeling, imitation, mixture, mixing, fibrous materials

The use of recurrent neural networks to predict the water level in the Amur River are consider. The advantages of using such networks in comparison with traditional machine learning methods are described. Various architectures of recurrent networks are compared, and hyperparameters of the model are optimized. The developed model based on long-term short-term memory (LSTM) has demonstrated high prediction accuracy, surpassing traditional methods. The results obtained can be used to improve the effectiveness of monitoring water resources and flood prevention.

Keywords: time series analysis, Amur, water level, forecasting, neural networks, recurrent network

The relevance of the research is determined by the need to solve complex optimization problems under conditions of high dimensionality, noisy data, and dynamically changing environments. Classical methods, such as genetic algorithms, often encounter the problem of premature convergence and fail to effectively adapt to changes in the problem. Therefore, this article focuses on identifying opportunities to enhance the flexibility and efficiency of evolutionary algorithms through integration with artificial neural networks, which allow for dynamically adjusting search parameters during the evolutionary process. The leading approach to addressing this problem is the development of a hybrid system that combines genetic algorithms with neural networks. This approach enables the neural network to adaptively regulate mutation and crossover probabilities based on the analysis of the current state of the population, preventing premature convergence and accelerating the search for the global extremum. The article presents methods for dynamic adjustment of evolutionary parameters using a neural network approach, reveals the principles of the hybrid system's operation, and provides results from testing on the Rastrigin function. The materials of the article hold practical value for further research in the field of optimization, particularly in solving problems with many local minima, where traditional methods may be ineffective. The application of the proposed hybrid model opens new perspectives for developing adaptive algorithms that can be used in various fields of science and engineering, where high accuracy and robustness to environmental changes are required.

Keywords: genetic algorithm, artificial neural network, dynamic tuning, hybrid method, global optimization, adaptive algorithm

The article explores the use of computer vision technologies to automate the process of observing animals in open spaces, with the aim of counting and identifying species. It discusses advanced methods of animal detection and recognition through the use of highly accurate neural networks. A significant challenge addressed in the study is the issue of duplicate animal counts in image data. To overcome this, two approaches are proposed: the analysis of video data sequences and the individual recognition of animals. The advantages and limitations of each method are analyzed in detail, alongside the potential benefits of combining both techniques to enhance the system's accuracy. The study also describes the process of training a neural network using a specialized dataset. Particular attention is given to the steps involved in data preparation, augmentation, and the application of neural networks like YOLO for efficient detection and classification. Testing results highlight the system's success in detecting animals, even under challenging conditions. Moreover, the article emphasizes the practical applications and potential of these technologies in monitoring animal populations and improving livestock management. It is noted that these advancements could contribute significantly to the development of similar systems in agriculture. The integration of such technologies is presented as a promising solution for tracking animal movement, assessing their health, and minimizing livestock losses across vast grazing areas.

Keywords: algorithm, computer vision, monitoring, pasture-based, livestock farming

This article will present the mlreflect package, written in Python, which is an optimized data pipeline for automated analysis of reflectometry data using machine learning. This package combines several methods of training and data processing. The predictions made by the neural network are accurate and reliable enough to serve as good starting parameters for subsequent data fitting using the least-mean-squares (LSC) method. For a large dataset consisting of 250 reflectivity curves of various thin films on silicon substrates, it was demonstrated that the analytical data pipeline with high accuracy finds the minimum of the film, which is very close to the set by the researcher using physical knowledge and carefully selected boundary conditions.

Keywords: neural network, radiography, thin films, data pipeline, machine learning

This paper is devoted to the application of the Winograd method to perform the wavelet transform in the problem of image compression. The application of this method reduces the computational complexity and also increases the speed of computation due to group processing of pixels. In this paper, the minimum number of bits at which high quality of processed images is achieved as a result of performing discrete wavelet transform in fixed-point computation format is determined. The experimental results showed that for processing fragments of 2 and 3 pixels without loss of accuracy using the Winograd method it is enough to use 2 binary decimal places for calculations. To obtain a high-quality image when processing groups of 4 and 5 pixels, it is sufficient to use 4 and 7 binary decimal places, respectively. Development of hardware accelerators of the proposed method of image compression is a promising direction for further research.

Keywords: wavelet transform, Winograd method, image processing, digital filtering, convolution with step

This paper presents the results of a study aimed at developing a method for semantic segmentation of thermal images using a modified neural network algorithm that differs from the original neural network algorithm by a higher speed or processing graphic information. As part of the study, a modification of the DeepLabv3+ semantic segmentation neural network algorithm was carried out by reducing the number of parameters of the neural network model, which made it possible to increase the speed of processing graphic information by 48% – from 27 to 40 frames per second. A training method is also presented that allows to increase the accuracy of the modified neural network algorithm; the accuracy value obtained was 5% lower than the accuracy of the original neural network algorithm.

Keywords: neural network algorithms, semantic segmentation, machine learning, data augmentation

This study presents a method for recognizing and classifying micro-expressions using optical flow analysis and the YOLOv11 architecture. Unlike previous binary detection approaches, this research enables multi-class classification while considering gender differences, as facial expressions may vary between males and females. A novel optical flow algorithm and a discretization technique improve classification stability, while the Micro ROC-AUC metric addresses class imbalance. Experimental results show that the proposed method achieves competitive accuracy, with gender-specific models further enhancing performance. Future work will explore ethnic variations and advanced learning strategies for improved recognition.

Keywords: microexpressions, pattern recognition, optical flow, YOLOv11

This article discusses the basic concepts and practical aspects of programming using the actor model on the Akka platform. The actor model is a powerful tool for creating parallel and distributed systems, providing high performance, fault tolerance and scalability. The article describes in detail the basic principles of how actors work, their lifecycle, and messaging mechanisms, as well as provides examples of typical patterns such as Master/Worker and Proxy. Special attention is paid to clustering and remote interaction of actors, which makes the article useful for developers working on distributed systems.

Keywords: actor model, akka, parallel programming, distributed systems, messaging, clustering, fault tolerance, actor lifecycle, programming patterns, master worker, proxy actor, synchronization, asynchrony, scalability, error handling

Mathematical modeling, numerical methods and program complexes (technical sciences). Geopolitical situation analysis of a number of episodes of the American Revolution in the context of applying structural balance and mathematical modeling methods. Structural balance management can help to find the most optimal strategies for interacting parties. This approach is used in a set of disciplines. In this article, the author analyzes examples of actors' interaction in the context of the American Revolution, which allows us to evaluate the state of affairs at this historical stage in an illustrative form. This approach is universal and is able to emphasize the management of structural balance in systems with actors, each of which has its own features and interests. A number of specific historical episodes serves as an example of the balanced and unbalanced systems. Each episode has its explanation in the frame of history. During the American Revolution, actors (countries and specific politicians, as well as indigenous peoples) had their own goals and interests, and their positive or negative interactions shaped the course of history in many ways.

Keywords: mathematical modeling, structural balance, discrete models, sign graph, U.S. history

The article is devoted to the development of a tool for automated generation of time constraints in the context of circuit development in the basis of programmable logic integrated circuits (FPGAs). The paper analyzes current solutions in the field of interface tools for generating design constraints. The data structure for the means of generating design constraints and algorithms for reading and writing Synopsys Design Constraints format files have been developed. Based on the developed structures and algorithms, a software module was implemented, which was subsequently implemented into the circuit design flow in the FPGA basis - X-CAD.

Keywords: computer-aided design, field programmable gate array, automation, design constraints, development, design route, interface, algorithm, tool, static timing analysis

The article presents an analysis of the application of the Socratic method for selecting machine learning models in corporate information systems. The study aims to explore the potential of utilizing the modular architecture of Socratic Models for integrating pretrained models without the need for additional training. The methodology relies on linguistic interactions between modules, enabling the combination of data from various domains, including text, images, and audio, to address multimodal tasks. The results demonstrate that the proposed approach holds significant potential for optimizing model selection, accelerating decision-making processes, and reducing the costs associated with implementing artificial intelligence in corporate environments.

Keywords: Socratic method, machine learning, corporate information systems, multimodal data, linguistic interaction, business process optimization, artificial intelligence