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.

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.

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.

In case of vessel wall damage or contact of blood plasma with a foreign surface, the chain of chemical reactions called coagulation cascade is launched that leading to the formation of a fibrin clot. A key enzyme of the coagulation cascade is thrombin, which catalyzes formation of fibrin from fibrinogen. The distribution of thrombin concentration in blood plasma determines spatio-temporal dynamics of clot formation. Contact pathway of blood coagulation triggers the production of thrombin in response to the contact with a negatively charged surface. If the concentration of thrombin generated at this stage is large enough, further production of thrombin takes place due to positive feedback loops of the coagulation cascade. As a result, thrombin propagates in plasma cleaving fibrinogen that results in the clot formation. The concentration profile and the speed of propagation of thrombin are constant and do not depend on the type of the initial activator.

Such behavior of the coagulation system is well described by the traveling wave solutions in a system of “reaction – diffusion” equations on the concentration of blood factors involved in the coagulation cascade. In this study, we carried out detailed analysis of the mathematical model describing the main reaction of the intrinsic pathway of coagulation cascade.We formulate necessary and sufficient conditions of the existence of the traveling wave solutions. For the considered model the existence of such solutions is equivalent to the existence of the wave solutions in the simplified one-equation model describing the dynamics of thrombin concentration derived under the quasi-stationary approximation.

Simplified model also allows us to obtain analytical estimate of the thrombin propagation rate in the considered model. The speed of the traveling wave for one equation is estimated using the narrow reaction zone method and piecewise linear approximation. The resulting formulas give a good approximation of the velocity of propagation of thrombin in the simplified, as well as in the original model.

The biological structure is considered as an open nonequilibrium system which properties can be described on the basis of kinetic equations. New problems with nonequilibrium boundary conditions are introduced. The nonequilibrium distribution tends gradually to an equilibrium state. The region of spatial inhomogeneity has a scale depending on the rate of mass transfer in the open system and the characteristic time of metabolism. In the proposed approximation, the internal energy of the motion of molecules is much less than the energy of translational motion. Or in other terms we can state that the kinetic energy of the average blood velocity is substantially higher than the energy of chaotic motion of the same particles. We state that the relaxation problem models a living system. The flow of entropy to the system decreases in downstream, this corresponds to Shrödinger’s general ideas that the living system “feeds on” negentropy. We introduce a quantity that determines the complexity of the biosystem, more precisely, this is the difference between the nonequilibrium kinetic entropy and the equilibrium entropy at each spatial point integrated over the entire spatial region. Solutions to the problems of spatial relaxation allow us to estimate the size of biosystems as regions of nonequilibrium. The results are compared with empirical data, in particular, for mammals we conclude that the larger the size of animals, the smaller the specific energy of metabolism. This feature is reproduced in our model since the span of the nonequilibrium region is larger in the system where the reaction rate is shorter, or in terms of the kinetic approach, the longer the relaxation time of the interaction between the molecules. The approach is also used for estimation of a part of a living system, namely a green leaf. The problems of aging as degradation of an open nonequilibrium system are considered. The analogy is related to the structure, namely, for a closed system, the equilibrium of the structure is attained for the same molecules while in the open system, a transition occurs to the equilibrium of different particles, which change due to metabolism. Two essentially different time scales are distinguished, the ratio of which is approximately constant for various animal species. Under the assumption of the existence of these two time scales the kinetic equation splits in two equations, describing the metabolic (stationary) and “degradative” (nonstationary) parts of the process.

In this work, we consider a special approximation of the general perturbation formula for the electromagnetic field by a set of electrically small inhomogeneities located in the domain of interest. The problem considered in this paper arises in many applications of technical electrodynamics, radar technologies and subsurface remote sensing. In the general case, it is formulated as follows: at some point in the perturbed domain, it is necessary to determine the amplitude of the electromagnetic field. The perturbation of electromagnetic waves is caused by a set of electrically small scatterers distributed in space. The source of electromagnetic waves is also located in perturbed domain. The problem is solved by introducing the far field approximation and through the formulation for the scatterer radar cross section value. This, in turn, allows one to significantly speed up the calculation process of the perturbed electromagnetic field by a set of a spherical inhomogeneities identical to each other with arbitrary electrophysical parameters. In this paper, we consider only the direct scattering problem; therefore, all parameters of the scatterers are known. In this context, it may be argued that the formulation corresponds to the well-posed problem and does not imply the solution of the integral equation in the generalized formula. One of the features of the proposed algorithm is the allocation of a characteristic plane at the domain boundary. All points of observation of the state of the system belong to this plane. Set of the scatterers is located inside the observation region, which is formed by this surface. The approximation is tested by comparing the results obtained with the solution of the general formula method for the perturbation of the electromagnetic field. This approach, among other things, allows one to remove a number of restrictions on the general perturbation formula for E-filed analysis.

Rugose spiraling whitefly (RSW) is one of the major pests which affects the coconut trees. It feeds on the tree by sucking up the water content as well as the essential nutrients from leaves. It also forms sooty mold in leaves due to which the process of photosynthesis is inhibited. Biocontrol of pest is harmless for trees and crops. The experimental results in literature reveal that Pseudomallada astur is a potential predator for this pest. We investigate the dynamics of predator, Pseudomallada astur’s interaction with rugose spiralling whitefly, Aleurodicus rugioperculatus in coconut trees using a mathematical model. In this system of ordinary differential equation, the pest-predator interaction is modeled using Holling type III functional response. The parametric values are calculated from the experimental results and are tabulated. An approximate analytical solution for the system has been derived. The homotopy analysis method proves to be a suitable method for creating solutions that are valid even for moderate to large parameter values, hence we employ the same to solve this nonlinear model. The $\hbar$-curves, which give the admissible region of $\hbar$, are provided to validate the region of convergence. We have derived the approximate solution at fifth order and stopped at this order since we obtain a more approximate solution in this iteration. Numerical simulation is obtained through MATLAB. The analytical results are compared with numerical simulation and are found to be in good agreement. The biological interpretation of figures implies that the use of a predator reduces the whitefly’s growth to a greater extent.

In this paper, we consider two variants of the concept of sharp minimum for mathematical programming problems with quasiconvex objective function and inequality constraints. It investigated the problem of describing a variant of a simple subgradient method with switching along productive and non-productive steps, for which, on a class of problems with Lipschitz functions, it would be possible to guarantee convergence with the rate of geometric progression to the set of exact solutions or its vicinity. It is important that to implement the proposed method there is no need to know the sharp minimum parameter, which is usually difficult to estimate in practice. To overcome this problem, the authors propose to use a step adjustment procedure similar to that previously proposed by B. T. Polyak. However, in this case, in comparison with the class of problems without constraints, it arises the problem of knowing the exact minimal value of the objective function. The paper describes the conditions for the inexactness of this information, which make it possible to preserve convergence with the rate of geometric progression in the vicinity of the set of minimum points of the problem. Two analogs of the concept of a sharp minimum for problems with inequality constraints are considered. In the first one, the problem of approximation to the exact solution arises only to a pre-selected level of accuracy, for this, it is considered the case when the minimal value of the objective function is unknown; instead, it is given some approximation of this value. We describe conditions on the inexact minimal value of the objective function, under which convergence to the vicinity of the desired set of points with a rate of geometric progression is still preserved. The second considered variant of the sharp minimum does not depend on the desired accuracy of the problem. For this, we propose a slightly different way of checking whether the step is productive, which allows us to guarantee the convergence of the method to the exact solution with the rate of geometric progression in the case of exact information. Convergence estimates are proved under conditions of weak convexity of the constraints and some restrictions on the choice of the initial point, and a corollary is formulated for the convex case when the need for an additional assumption on the choice of the initial point disappears. For both approaches, it has been proven that the distance from the current point to the set of solutions decreases with increasing number of iterations. This, in particular, makes it possible to limit the requirements for the properties of the used functions (Lipschitz-continuous, sharp minimum) only for a bounded set. Some computational experiments are performed, including for the truss topology design problem.

The study of elastic waves in porous media is relevant for mineral exploration, the use of porous screens for shock wave damping, and the study of the structure of the earth’s crust. The elastic properties of a porous medium, which can be judged by the propagation velocity of various types of waves, depend on the degree of consolidation of the porous medium. For example, bulk media (sand, glass beads, granular materials) have a low sound velocity (about 100 m/s); compaction of such media is accompanied by a slight increase in velocity, while their consolidation (sandstone, gas hydrate cementation) leads to a multiple increase in the acoustic wave velocity, on the order of 2000–3000 m/s. This paper theoretically investigates the dynamics of a wave pulse in a shock tube containing a layer of a bulk medium. Numerical modeling was performed under experimental conditions. A description of a shock tube experimental setup is provided. The setup consists of a high-pressure volume (HPV), a low-pressure volume (LPV), and a bulk medium section. A shock wave pulse (SWP) is generated by the rupture of a diaphragm between the HPV and LPV. The SWP dynamics are recorded by piezoelectric sensors located flush on the inside of the tube. In the shock tube, equipped with a bulk medium section, the wave experiences multiple reflections from the surface of the porous medium under study and the upper end of the tube. The reflected signals are used as probe pulses to study changes in the porous medium caused by repeated passages of the shock wave pulse, with a period of approximately 10 ms. A mathematical model is used that includes the equations of conservation of mass, momentum, and energy for the gas phase and solid particles with closure relations. The process is described for one-dimensional planar motion of the gas and dispersed phases. The numerical solution utilizes an approximation of the equations based on the control volume method. Numerical results have shown that the proposed model accurately describes, qualitatively and quantitatively, the occurrence of a sharp, short-term increase in the total voltage (peak) during repeated pulse passage through a layer of bulk material, as observed in experiments.

The paper develops a new mathematical method of the joint signal and noise calculation at the Rice statistical distribution based on combing the maximum likelihood method and the method of moments. The calculation of the sough-for values of signal and noise is implemented by processing the sampled measurements of the analyzed Rician signal’s amplitude. The explicit equations’ system has been obtained for required signal and noise parameters and the results of its numerical solution are provided confirming the efficiency of the proposed technique. It has been shown that solving the two-parameter task by means of the proposed technique does not lead to the increase of the volume of demanded calculative resources if compared with solving the task in one-parameter approximation. An analytical solution of the task has been obtained for the particular case of small value of the signal-to-noise ratio. The paper presents the investigation of the dependence of the sought for parameters estimation accuracy and dispersion on the quantity of measurements in experimental sample. According to the results of numerical experiments, the dispersion values of the estimated sought-for signal and noise parameters calculated by means of the proposed technique change in inverse proportion to the quantity of measurements in a sample. There has been implemented a comparison of the accuracy of the soughtfor Rician parameters’ estimation by means of the proposed technique and by earlier developed version of the method of moments. The problem having been considered in the paper is meaningful for the purposes of Rician data processing, in particular, at the systems of magnetic-resonance visualization, in devices of ultrasonic visualization, at optical signals’ analysis in range-measuring systems, at radar signals’ analysis, as well as at solving many other scientific and applied tasks that are adequately described by the Rice statistical model.