All issues

The paper presents a hybrid artificial neural network model based on an architecture that incorporates a convolutional image encoder (CNN) and an attention module (Attention-based Multiple Instance Learning, Attention MIL). This module aggregates informative features from a sequence of frames capturing the flame spraying process. Additional technological parameters—air pressure, propane pressure, and standoff distance — are integrated into the model via a tabular channel, enabling it to account for the relationship between visual data and numerical process regime characteristics. The software implementation was developed using the Streamlit platform and the PyTorch library. It features an interactive interface for model training and result visualization, analysis of attention weights across frames, and a prediction mode for output characteristics: surface roughness ($R_a$) and the mass of the deposited coating ($m$). Experimental studies were conducted on data from real-world technological processes, and a comparative analysis of the accuracy of various model configurations was performed. The results demonstrate that the hybrid neural network, which combines visual and tabular features, achieves higher prediction accuracy compared to models using only a single modality. Furthermore, when comparing different implementations of the hybrid network, it was established that using the attention mechanism to process the series of flame spray images provides a significant increase in accuracy over a simple averaging of features without attention. The application includes an attention visualization module that creates a montage of the most significant frames and displays their attention weights, allowing users to identify which frames had the greatest influence on the prediction. The model’s capability for export to the ONNX format for integration into process control systems is also demonstrated. The proposed approach showcases the effectiveness of fusing visual and tabular information for manufacturing process monitoring tasks. The model can serve as a foundation for developing a decision support system or an automated quality control system for coatings produced by flame spraying. The limitations of the implemented model and prospects for its further development are also considered.

The article describes the state of the art computer simulation in the field of metal forming processes, the main problem points of traditional methods were identified. The method, that allows to predict the deformation distribution in the volume of deformable metal with taking into account of microstructure behavioral characteristics in deformation load conditions, was described. The method for optimizing computational resources of multiscale models by using statistical similar representative volume elements (SSRVE) was presented. The modeling methods were tested on the process of single pass drawing of round rod from steel grade 20. In a comparative analysis of macro and micro levels models differences in quantitative terms of the stress-strain state and their local distribution have been identified. Microlevel model also allowed to detect the compressive stresses and strains, which were absent at the macro level model. Applying the SSRVE concept repeatedly lowered the calculation time of the model while maintaining the overall accuracy.

This paper is devoted to the analysis of the effect of additive and parametric noise on the processes occurring in the nerve cell. This study is carried out on the example of the well-known Morris – Lecar model described by the two-dimensional system of ordinary differential equations. One of the main properties of the neuron is the excitability, i.e., the ability to respond to external stimuli with an abrupt change of the electric potential on the cell membrane. This article considers a set of parameters, wherein the model exhibits the class 2 excitability. The dynamics of the system is studied under variation of the external current parameter. We consider two parametric zones: the monostability zone, where a stable equilibrium is the only attractor of the deterministic system, and the bistability zone, characterized by the coexistence of a stable equilibrium and a limit cycle. We show that in both cases random disturbances result in the phenomenon of the stochastic generation of mixed-mode oscillations (i. e., alternating oscillations of small and large amplitudes). In the monostability zone this phenomenon is associated with a high excitability of the system, while in the bistability zone, it occurs due to noise-induced transitions between attractors. This phenomenon is confirmed by changes of probability density functions for distribution of random trajectories, power spectral densities and interspike intervals statistics. The action of additive and parametric noise is compared. We show that under the parametric noise, the stochastic generation of mixed-mode oscillations is observed at lower intensities than under the additive noise. For the quantitative analysis of these stochastic phenomena we propose and apply an approach based on the stochastic sensitivity function technique and the method of confidence domains. In the case of a stable equilibrium, this confidence domain is an ellipse. For the stable limit cycle, this domain is a confidence band. The study of the mutual location of confidence bands and the boundary separating the basins of attraction for different noise intensities allows us to predict the emergence of noise-induced transitions. The effectiveness of this analytical approach is confirmed by the good agreement of theoretical estimations with results of direct numerical simulations.

The article proposes a method of joint analysis of multidimensional financial time series based on the evaluation of the set of properties of stock quotes in a sliding time window and the subsequent averaging of property values for all analyzed companies. The main purpose of the analysis is to construct measures of joint behavior of time series reacting to the occurrence of a synchronous or coherent component. The coherence of the behavior of the characteristics of a complex system is an important feature that makes it possible to evaluate the approach of the system to sharp changes in its state. The basis for the search for precursors of sharp changes is the general idea of increasing the correlation of random fluctuations of the system parameters as it approaches the critical state. The increments in time series of stock values have a pronounced chaotic character and have a large amplitude of individual noises, against which a weak common signal can be detected only on the basis of its correlation in different scalar components of a multidimensional time series. It is known that classical methods of analysis based on the use of correlations between neighboring samples are ineffective in the processing of financial time series, since from the point of view of the correlation theory of random processes, increments in the value of shares formally have all the attributes of white noise (in particular, the “flat spectrum” and “delta-shaped” autocorrelation function). In connection with this, it is proposed to go from analyzing the initial signals to examining the sequences of their nonlinear properties calculated in time fragments of small length. As such properties, the entropy of the wavelet coefficients is used in the decomposition into the Daubechies basis, the multifractal parameters and the autoregressive measure of signal nonstationarity. Measures of synchronous behavior of time series properties in a sliding time window are constructed using the principal component method, moduli values of all pairwise correlation coefficients, and a multiple spectral coherence measure that is a generalization of the quadratic coherence spectrum between two signals. The shares of 16 large Russian companies from the beginning of 2010 to the end of 2016 were studied. Using the proposed method, two synchronization time intervals of the Russian stock market were identified: from mid-December 2013 to mid- March 2014 and from mid-October 2014 to mid-January 2016.

The development of a viral infection in the organism is a complex process which depends on the competition race between virus replication in the host cells and the immune response. To study different regimes of infection progression, we analyze the general mathematical model of immune response to viral infection. The model consists of two ODEs for virus and immune cells non-dimensionalized concentrations. The proliferation rate of immune cells in the model is represented by a bell-shaped function of the virus concentration. This function increases for small virus concentrations describing the antigen-stimulated clonal expansion of immune cells, and decreases for sufficiently high virus concentrations describing down-regulation of immune cells proliferation by the infection. Depending on the virus virulence, strength of the immune response, and the initial viral load, the model predicts several scenarios: (a) infection can be completely eliminated, (b) it can remain at a low level while the concentration of immune cells is high; (c) immune cells can be essentially exhausted, or (d) completely exhausted, which is accompanied (c, d) by high virus concentration. The analysis of the model shows that virus concentration can oscillate as it gradually converges to its equilibrium value. We show that the considered model can be obtained by the reduction of a more general model with an additional equation for the total viral load provided that this equation is fast. In the case of slow kinetics of the total viral load, this more general model should be used.

The COVID-19 pandemic has caused increased interest in mathematical models of the epidemic process, since only statistical analysis of morbidity does not allow medium-term forecasting in a rapidly changing situation.

Among the specific features of COVID-19 that need to be taken into account in mathematical models are the heterogeneity of the pathogen, repeated changes in the dominant variant of SARS-CoV-2, and the relative short duration of post-infectious immunity.

In this regard, solutions to a system of differential equations for a SIR class model with a heterogeneous duration of post-infectious immunity were analytically studied, and numerical calculations were carried out for the dynamics of the system with an average duration of post-infectious immunity of the order of a year.

For a SIR class model with a heterogeneous duration of post-infectious immunity, it was proven that any solution can be continued indefinitely in time in a positive direction without leaving the domain of definition of the system.

For the contact number $R_0 \leqslant 1$, all solutions tend to a single trivial stationary solution with a zero share of infected people, and for $R_0 > 1$, in addition to the trivial solution, there is also a non-trivial stationary solution with non-zero shares of infected and susceptible people. The existence and uniqueness of a non-trivial stationary solution for $R_0 > 1$ was proven, and it was also proven that it is a global attractor.

Also, for several variants of heterogeneity, the eigenvalues of the rate of exponential convergence of small deviations from a nontrivial stationary solution were calculated.

It was found that for contact number values corresponding to COVID-19, the phase trajectory has the form of a twisting spiral with a period length of the order of a year.

This corresponds to the real dynamics of the incidence of COVID-19, in which, after several months of increasing incidence, a period of falling begins. At the same time, a second wave of incidence of a smaller amplitude, as predicted by the model, was not observed, since during 2020–2023, approximately every six months, a new variant of SARS-CoV-2 appeared, which was more infectious than the previous one, as a result of which the new variant replaced the previous one and became dominant.

Deep learning’s power stems from complex architectures; however, these can lead to overfitting, where models memorize training data and fail to generalize to unseen examples. This paper proposes a novel probabilistic approach to mitigate this issue. We introduce two key elements: Truncated Log-Uniform Prior and Truncated Log-Normal Variational Approximation, and Automatic Relevance Determination (ARD) with Bayesian Deep Neural Networks (BDNNs). Within the probabilistic framework, we employ a specially designed truncated log-uniform prior for noise. This prior acts as a regularizer, guiding the learning process towards simpler solutions and reducing overfitting. Additionally, a truncated log-normal variational approximation is used for efficient handling of the complex probability distributions inherent in deep learning models. ARD automatically identifies and removes irrelevant features or weights within a model. By integrating ARD with BDNNs, where weights have a probability distribution, we achieve a variational bound similar to the popular variational dropout technique. Dropout randomly drops neurons during training, encouraging the model not to rely heavily on any single feature. Our approach with ARD achieves similar benefits without the randomness of dropout, potentially leading to more stable training.

To evaluate our approach, we have tested the model on two datasets: the Canadian Institute For Advanced Research (CIFAR-10) for image classification and a dataset of Macroscopic Images of Wood, which is compiled from multiple macroscopic images of wood datasets. Our method is applied to established architectures like Visual Geometry Group (VGG) and Residual Network (ResNet). The results demonstrate significant improvements. The model reduced overfitting while maintaining, or even improving, the accuracy of the network’s predictions on classification tasks. This validates the effectiveness of our approach in enhancing the performance and generalization capabilities of deep learning models.

For cognitive radio systems, it is important to use efficient algorithms that search for free channels that can be provided to secondary users. Therefore, this paper is devoted to improving the accuracy of prediction frequency resource occupancy of a cellular communication system using spatiotemporal radio environment maps. The formation of a radio environment map is implemented for the fourthgeneration cellular communication system Long-Term Evolution. Taking this into account, a model structure has been developed that includes data generation and allows training and testing of an artificial neural network to predict the occupancy of frequency resources presented as the contents of radio environment map cells. A method for assessing prediction accuracy is described. The simulation model of the cellular communication system is implemented in the MatLab. The developed frequency resource occupancy prediction model is implemented in the Python. The complete file structure of the model is presented. The experiments were performed using artificial neural networks based on the Long Short-Term Memory and Kolmogorov – Arnold neural network architectures, taking into account its modification. It was found that with an equal number of parameters, the Kolmogorov –Arnold neural network learns faster for a given task. The obtained research results indicate an increase in the accuracy of prediction the occupancy of the frequency resource of the cellular communication system when using the Kolmogorov – Arnold neural network.

The paper presents a cascade model of a physically supported neural network designed to predict pressure drop in three-phase flow (oil, gas, water) in a pipe section with various angles of inclination. To overcome the constraints of existing empirical correlations and computation-intensive numerical modeling methods, we propose an architecture that decomposes the problem into three sequential physically interpretable subtasks: regression prediction of the fluid hold-up coefficient, fluid flow regime classification, and pressure gradient evaluation. Each subtask is solved by a separate fully connected neural network, the output of which is passed to the next model in the cascade. Training and testing of the proposed architecture was performed on an extensive synthetic dataset (8 · 10⁷ records) generated using a semi-empirical model. Verification is performed on independent experimental data. A comparative analysis with a single fully connected (non-cascade) neural network is made, and the sensitivity of the models is examined using Sobol and Borgonovo methods. The cascade model demonstrates superior accuracy and ensures high interpretability of results by providing intermediate physical parameters (fluid hold-up coefficient, flow regime). The developed model has low computational complexity, which allows it to be used in real-time systems and digital twins of hydraulic systems in the oil and gas industry.

This paper formulates and solves a practical problem of data recovery regarding the distribution of languages on regional level in context of China. The necessity of this recovery is related to the problem of the determination of the linguistic diversity indices, which, in turn, are used to analyze empirically and to predict sources of social and economic development as well as to indicate potential conflicts at regional level. We use Ethnologue database and China census as the initial data sources. For every language spoken in China, the data contains (a) an estimate of China residents who claim this language to be their mother tongue, and (b) indicators of the presence of such residents in China provinces. For each pair language/province, we aim to estimate the number of the province inhabitants that claim the language to be their mother tongue. This base problem is reduced to solving an undetermined system of algebraic equations. Given additional restriction that Ethnologue database introduces data collected at different time moments because of gaps in Ethnologue language surveys and accompanying data collection expenses, we relate those data to a single time moment, that turns the initial task to an ’ill-posed’ system of algebraic equations with imprecisely determined right hand side. Therefore, we are looking for an approximate solution characterized by a minimal discrepancy of the system. Since some languages are much less distributed than the others, we minimize the weighted discrepancy, introducing weights that are inverse to the right hand side elements of the equations. This definition of discrepancy allows to recover the required variables. More than 92% of the recovered variables are robust to probabilistic modelling procedure for potential errors in initial data.