All issues

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.

Modeling of channel processes in the study of coastal channel deformations requires the calculation of hydrodynamic flow parameters that take into account the existence of secondary transverse currents formed at channel curvature. Three-dimensional modeling of such processes is currently possible only for small model channels; for real river flows, reduced-dimensional models are needed. At the same time, the reduction of the problem from a three-dimensional model of the river flow movement to a two-dimensional flow model in the cross-section assumes that the hydrodynamic flow under consideration is quasi-stationary and the hypotheses about the asymptotic behavior of the flow along the flow coordinate of the cross-section are fulfilled for it. Taking into account these restrictions, a mathematical model of the problem of the a stationary turbulent calm river flow movement in a channel cross-section is formulated. The problem is formulated in a mixed formulation of velocity — “vortex – stream function”. As additional conditions for problem reducing, it is necessary to specify boundary conditions on the flow free surface for the velocity field, determined in the normal and tangential direction to the cross-section axis. It is assumed that the values of these velocities should be determined from the solution of auxiliary problems or obtained from field or experimental measurement data.

To solve the formulated problem, the finite element method in the Petrov – Galerkin formulation is used. Discrete analogue of the problem is obtained and an algorithm for solving it is proposed. Numerical studies have shown that, in general, the results obtained are in good agreement with known experimental data. The authors associate the obtained errors with the need to more accurately determine the circulation velocities field at crosssection of the flow by selecting and calibrating a more appropriate model for calculating turbulent viscosity and boundary conditions at the free boundary of the cross-section.

Considered the application of physics-informed neural networks using multi layer perceptrons to solve Cauchy initial value problems in which the right-hand sides of the equation are continuous monotonically increasing, decreasing or oscillating functions. With the use of the computational experiments the influence of the construction of the approximate neural network solution, neural network structure, optimization algorithm and software implementation means on the learning process and the accuracy of the obtained solution is studied. The analysis of the efficiency of the most frequently used machine learning frameworks in software development with the programming languages Python and C# is carried out. It is shown that the use of C# language allows to reduce the time of neural networks training by 20–40%. The choice of different activation functions affects the learning process and the accuracy of the approximate solution. The most effective functions in the considered problems are sigmoid and hyperbolic tangent. The minimum of the loss function is achieved at the certain number of neurons of the hidden layer of a single-layer neural network for a fixed training time of the neural network model. It’s also mentioned that the complication of the network structure increasing the number of neurons does not improve the training results. At the same time, the size of the grid step between the points of the training sample, providing a minimum of the loss function, is almost the same for the considered Cauchy problems. Training single-layer neural networks, the Adam method and its modifications are the most effective to solve the optimization problems. Additionally, the application of twoand three-layer neural networks is considered. It is shown that in these cases it is reasonable to use the LBFGS algorithm, which, in comparison with the Adam method, in some cases requires much shorter training time achieving the same solution accuracy. The specificity of neural network training for Cauchy problems in which the solution is an oscillating function with monotonically decreasing amplitude is also investigated. For these problems, it is necessary to construct a neural network solution with variable weight coefficient rather than with constant one, which improves the solution in the grid cells located near by the end point of the solution interval.

The paper constructs and studies a generalized model describing two-component systems of reaction – diffusion type with power nonlinearity, considering the influence of external noise. A methodology has been developed for analyzing the generalized model, which includes linear stability analysis, nonlinear stability analysis, and numerical simulation of the system’s evolution. The linear analysis technique uses basic approaches, in which the characteristic equation is obtained using a linearization matrix. Nonlinear stability analysis realized up to third-order moments inclusively. For this, the functions describing the dynamics of the components are expanded in Taylor series up to third-order terms. Then, using the Novikov theorem, the averaging procedure is carried out. As a result, the obtained equations form an infinite hierarchically subordinate structure, which must be truncated at some point. To achieve this, contributions from terms higher than the third order are neglected in both the equations themselves and during the construction of the moment equations. The resulting equations form a set of linear equations, from which the stability matrix is constructed. This matrix has a rather complex structure, making it solvable only numerically. For the numerical study of the system’s evolution, the method of variable directions was chosen. Due to the presence of a stochastic component in the analyzed system, the method was modified such that random fields with a specified distribution and correlation function, responsible for the noise contribution to the overall nonlinearity, are generated across entire layers. The developed methodology was tested on the reaction – diffusion model proposed by Barrio et al., according to the results of the study, they showed the similarity of the obtained structures with the pigmentation of fish. This paper focuses on the system behavior analysis in the neighborhood of a non-zero stationary point. The dependence of the real part of the eigenvalues on the wavenumber has been examined. In the linear analysis, a range of wavenumber values is identified in which Turing instability occurs. Nonlinear analysis and numerical simulation of the system’s evolution are conducted for model parameters that, in contrast, lie outside the Turing instability region. Nonlinear analysis found noise intensities of additive noise for which, despite the absence of conditions for the emergence of diffusion instability, the system transitions to an unstable state. The results of the numerical simulation of the evolution of the tested model demonstrate the process of forming spatial structures of Turing type under the influence of additive noise.

The article describes the migration process of a certain population, taking into account the spatial heterogeneity of the local capacity of the ecological niche. It is assumed that this spatial heterogeneity is caused by various natural or artificial factors. The mathematical model of the migration process under consideration is a Cauchy problem on a straight line for some quasi-linear partial differential equation of the first order, which is satisfied by the linear population density under consideration. In this paper, a general solution to this Cauchy problem is found for an arbitrary dependence of the local capacity of an ecological niche on the spatial coordinate. This general solution was applied to describe the migration of the population in question in two different cases: in the case of a dependence of the local capacity of the ecological niche on the spatial coordinate in the form of a smooth step and in the case of a hill-like dependence of the local capacity of the ecological niche on the spatial coordinate. In both cases, the solution to the Cauchy problem is expressed in terms of higher transcendental functions. By applying special relations to the model parameters, these higher transcendental functions are reduced to elementary functions, which makes it possible to obtain exact model solutions explicitly expressed in terms of elementary functions. With the help of these precise solutions, an extensive program of computational experiments has been implemented, showing how the initial population density of the Gaussian form is dispersed by the considered two types of spatial heterogeneity of the local capacity of the ecological niche. These computational experiments have shown that when passing through both step-like and hill-like spatial inhomogeneities of the local capacity of an ecological niche with a narrow Gaussian width of its initial density compared to the characteristic spatial scale of these inhomogeneities, the system forgets its initial state. In particular, if we interpret the system under study as a population living in an extended calm rectilinear river along its bed, then it can be argued that under this initial condition, after the current of this river carries the population under consideration through the area of spatial heterogeneity of the local capacity of the ecological niche, the population density becomes a quasi-rectangular function.

This paper considers the inverse problem of seismic exploration — determining the structure of the media based on the recorded wave response from it. Large cracks are considered as target objects, whose size and position are to be determined.

he direct problem is solved using the grid-characteristic method. The method allows using physically based algorithms for calculating outer boundaries of the region and contact boundaries inside the region. The crack is assumed to be thin, a special condition on the crack borders is used to describe the crack.

The inverse problem is solved using convolutional neural networks. The input data of the neural network are seismograms interpreted as images. The output data are masks describing the medium on a structured grid. Each element of such a grid belongs to one of two classes — either an element of a continuous geological massif, or an element through which a crack passes. This approach allows us to consider a medium with an unknown number of cracks.

The neural network is trained using only samples with one crack. The final testing of the trained network is performed using additional samples with several cracks. These samples are not involved in the training process. The purpose of testing under such conditions is to verify that the trained network has sufficient generality, recognizes signs of a crack in the signal, and does not suffer from overtraining on samples with a single crack in the media.

The paper shows that a convolutional network trained on samples with a single crack can be used to process data with multiple cracks. The networks detects fairly small cracks at great depths if they are sufficiently spatially separated from each other. In this case their wave responses are clearly distinguishable on the seismogram and can be interpreted by the neural network. If the cracks are close to each other, artifacts and interpretation errors may occur. This is due to the fact that on the seismogram the wave responses of close cracks merge. This cause the network to interpret several cracks located nearby as one. It should be noted that a similar error would most likely be made by a human during manual interpretation of the data. The paper provides examples of some such artifacts, distortions and recognition errors.

Laser interference microscopy was used to study morphology and intracellular dynamics of erythrocytes, neurons and mast cells. We have found that changes of the local refractive index (RI) of cells have regular components that relate to the cooperative processes in the cellular submembrane and centre regions. We have shown that characteristic frequencies of RI dynamics differ for various cell types and can be used as markers of specific cellular processes.

We consider two possible computational methods of the interpretation of experimental data obtained by means of the magnetic force microscopy. These methods of macrospin distribution simulation and reconstruction can be used for research of magnetization reversal processes of nanodots in ordered 2D arrays of nanodots. New approaches to the development of high-performance superscale algorithms for parallel executing on a supercomputer clusters for solving direct and inverse task of the modeling of magnetic states, types of ordering, reversal processes of nanosystems with a collective behavior are proposed. The simulation results are consistent with experimental results.

The dynamics of coupled elements’ ensembles are investigated in the context of description of spatio-temporal processes in the myocardium. Basic element is map-based model constructed by simplification and reduction of Luo-Rudy model. In particular, capabilities of the model in replication of different regimes of cardiac activity are shown, including excitable and oscillatory regimes. The dynamics of 1D and 2D lattices of coupled oscillatory elements with a random distribution of individual frequencies are considered. Effects of cluster synchronization and transition to global synchronization by increasing of coupling strength are discussed. Impulse propagation in the chain of excitable cells has been observed. Analysis of 2D lattice of excitable elements with target and spiral waves have been made. The characteristics of the spiral wave has been analyzed in depending on the individual parameters of the map and coupling strength between elements of the lattice. A study of mixed ensembles consisting of excitable and oscillatory elements with a gradient changing of the properties have been made, including the task for description of normal and pathological activity of the sinoatrial node.

This paper presents a model mixture of probability distributions of signal and noise. Typically, when analyzing the data under conditions of uncertainty it is necessary to use nonparametric tests. However, such an analysis of nonstationary data in the presence of uncertainty on the mean of the distribution and its parameters may be ineffective. The model involves the implementation of a case of a priori non-parametric uncertainty in the processing of the signal at a time when the separation of signal and noise are related to different general population, is feasible.