Information on the passage of photon radiation from photon sources in the energy range from 6 to 24 MeV in various protective materials with thicknesses in the range from 15 to 80 cm is reported in [7–10]. The use of electron accelerators in industry and medicine with the primary electron beam energy higher than these energies requires data on the characteristics of the attenuation of bremsstrahlung photons for energy higher than these energies and thickness larger than these thicknesses. Concrete, iron, and lead are used as protective materials against the bremsstrahlung of electron accelerators. In this work the energy distributions of photon flux densities and effective doses in planes of concrete protections, as well as similar characteristics of the fields of neutron and secondary photon radiation, have been calculated. The calculations have been performed for two types of sources: a flat unidirectional monoenergetic photon source with an energy of 30 MeV and a bremsstrahlung source with a maximum energy of 30 MeV.

   Any transportation of spent nuclear fuel is a complex multifactorial problem requiring the coordination of numerous technical means including specialized transport containers, fasteners, reloading devices, grippers, dampers, and special wagons. One of the ways to solve such problems is to test a prototype of the fuel package unit under conditions simulating an aviation accident (collision with a rigid barrier at a velocity of 90 m/s). Tests using the capabilities of a rocket track have been presented in this work. An aerodynamic flow around the section of a video recording unit has been calculated under the conditions of an aeroballistic experiment. The strength and natural frequencies of the recording instrument used in the aeroballistic experiment have been evaluated. The modal calculation of the recording instrument has been performed using the finite element method. Ten forms of free oscillations of the recording instrument have been obtained. It has been concluded that the instrument is performable. The calculations have allowed installing a high-speed camera on the stage of a rocket train to increase the information content in the experiment for obtaining frames of acceleration and separation of the fuel package unit. Photos of the acceleration and separation of the fuel packing kit from the rocket train taken by the aboard camera have been presented. The reported studies have made it possible to justify and apply high-speed shooting from the upper stage of the rocket train in the simulation of the aviation accident for testing transport packaging kits according to the IAEA standards.

   This paper presents the results of production and radiographic examination of micro-samples of curium-244 compounds with ion, cobalt and carbon, prepared by high temperature condensation of metal curium vapor onto corresponding substrates. In the Cm–Co system three intermetallic compounds were detected: Co₁₇Cm₂ (hexagonal lattice of spatial group P6₃/mcm), Co₅Cm (hexagonal lattice of spatial group P6/mmm) and intermetallide Co₂Cm (cubic lattice of spatial group Fd3m). In the Cm–Fe system two intermetallic compounds were detected: Fe₁₇Cm₂ (hexagonal lattice of spatial group P6₃/mcm ) and Fe₂Cm (cubic lattice of spatial group Fd3m). A comparison of the obtained intermetallic compounds in the Cm-Co and Cm–Fe systems with the result of studying the interaction of Cm-244 with nickel is given.

   A fourth-order nonlinear partial differential equation with the third- and fifth-degree nonlinearities has been considered. This equation can be used to describe pulses in optical fibers. The Cauchy problem for this equation cannot be solved by the inverse scattering transform method; for this reason, the equation is considered using the traveling wave variables. The substitution of a certain type solution gives a system of ordinary differential equations (ODEs) for the imaginary and real parts of the equation. The Painlevé test is applied to the resulting system of ODEs. According to the test, the considered ODE system does not have the Painlevé property because the expansion of the general solution into the Laurent series contains complex Fuchs indices. When using the Painlevé test, a condition for the velocity of a traveling wave at which the system is simplified to one ordinary fourth-order differential equation is obtained. The first integral is found for this equation. The method of the simplest equations is used to construct the exact solution of the considered ODE. The found solution has two arbitrary constants and is expressed in terms of the elliptic Weierstrass function. A special case where the solution has the form of a solitary wave is considered. Periodic and solitary wave solutions at different parameter values are illustrated.

   Various classes of nonlinear ordinary differential equations are considered. To construct exact solutions in an implicit form, the splitting method based on the generalized separation of variables is used. The main attention is paid to nonlinear equations of a sufficiently general form that contain one or more arbitrary functions. It is important to note that the exact solutions of nonlinear differential equations that depend on arbitrary functions and, therefore, are sufficiently general are of the greatest practical interest for testing numerical and approximate methods for solving various problems. Examples of particular nonlinear equations and their exact solutions are given. In some cases, it is possible to find general solutions of the equations or lower their order. The approach used can be generalized to nonlinear partial differential equations. New exact solutions with functional separation of variables are obtained for reaction-diffusion type equations.

   The propagation of solitary waves through optical media is described by a variety of nonlinear partial differential equations. In this work, the nonlinear dynamics of dispersive nonlinear waves in polarization–preserving fibers having Kerr nonlinearity has been described by the Radhakrishnan–Kundu–Lakshmanan equation. Dimensionless variables are used in the initial equation to study of dynamic processes. The introduction of the traveling wave variables reduces the equation to the system of two third order ordinary differential equations. The divergence is calculated for the normal form vector field of this system. It has been found that this system of equations is not dissipative. The largest Lyapunov exponents are computed by the Bennetin algorithm for the different values of one of the model parameters. Despite the presence of a positive largest Lyapunov exponent for some parameter values, there are no attractors or chaotic dynamical regimes in this system, since the solution unboundedly decreases in one variable and the initial point in the algorithm has to be chosen on the attractor.

   The paper considers the generative-discriminative model (GAN) as applied to the task of analyzing text data, in particular, determining the gender of the author of a Russian-language text. The approach developed belongs to semi-supervised learning algorithms, when both labeled and unlabeled samples are involved in the model fitting process. The GAN model is implemented as a deep neural network consisting of fully connected, recurrent and convolution layers. The basis of the generative part of the GAN model is a variational auto-encoder, which encodes the input sample into the space of hidden variables and then decodes the latter into the original representation. When decoding, the class label of the input example is used, known in the case of the labeled set or predicted by the classifier for unlabeled samples. The input for the model is a sequence of words, each encoded by a vector of the principal components of its morphological features. To provide the function of recovering texts with more than 50 words, the principles of the work of language models are used. The discriminant part is configured to determine whether a given sample was generated by an auto-encoder or taken from the original set. Quality assessment of the GAN model was carried out on a set of texts from LiveJournal blogs. It is shown that the use of the generative-discriminative model allows to improve the quality of classification by 2 % in the F1 metric, while reducing the standard deviation by 2–3 times when learning on a small number of labeled examples. Various modes of training and variations in the topology of the GAN model are investigated, and the most effective modes of operation of models of this type for the task of classifying texts are demonstrated.

   The adequacy of the reconstruction of pole figures (PFs) from the results of the electron backscatter diffraction (EBSD) experiment has been verified using a new technique for estimating the similarity of two PFs. This method is based on an analog of the Kolmogorov–Smirnov criterion on the S² sphere and takes into account the probabilistic nature of PFs. This method also proposes a quantitative similarity measure for the compared PFs. In this study, this method is utilized to quantitatively compare the PF reconstructed from the results of the EBSD experiment to the PF of a polycrystalline sample in order to estimate the effect of the selected values of the EBSD experiment parameters on the underlying distribution. Such a comparison becomes possible by mathematical modeling of a polycrystalline sample and a subsequent EBSD experiment. A new similarity measure between the sample PF and the PF obtained from the result of model EBSD measurements under the variation of the experimental parameters, namely, the scanning step and tolerance angle, has been calculated. The results have been compared to similar results for the RP factor widely used to estimate the differences between two PFs. Both metrics are utilized to study the effect of the texture sharpness parameter of the polycrystalline sample on the results of the EBSD experiment with various scanning steps and tolerance angles.

   Essential component of virtual tool is statistical one used for statistical models identification on qualification tests results, criteria models parameterization for forecasting at qualification tests, regression models synthesis, their optimization and adequateness analysis at factor planning. Regression model linearity allows controlling factors significance using standard statistical methods and performing model optimization using linear programming. At oils characteristics identification their non-normality was shown. To form rational processing algorithms some alternative approximations were formed. Bimodal density distribution was approximated with sum of different normal ones. One – modal distributions of other characteristics were approximated with Weibull distributions. Parameterization of these distributions was performed using moment’s method. For reproducibility estimation analytical Cochran distribution on the base of Fisher distribution and analytical Grabbs distribution on the base of Student distribution were developed. It was shown that preliminary normalization allows avoiding error decision about result reproducibility. Criteria model parameterization was proposed to perform on effect forecast at extremal conditions that allows to guaranty qualification normative. Preliminary choice of these conditions is realized by using step-by-step approach in the course of factor analysis. Developed method for similarity criteria synthesis is based on criterion logarithm presentation as linear form of some similarity numbers interpreted as independent factors. Coefficients of this form were estimated using factor analysis. Also the intuitive method of factors ranking in linear form using saturated plans and least significant factors rejection was developed. Its base advantage is fictive factors exclusion and base disadvantage – regular method absence for significance thresholds set.

   The possibility for a two-layer spiking neural network with spike-timing-dependent plasticity to be trained using the output rate stabilization effect is studied. Numerical simulations show that high correlation of input spike sequences disrupts the stabilization of the neuron spiking rate because of insufficient competition between input synapses. For a two-layer network, shown spikes coming at the input of the second layer can be correlated because the first-layer weights established by STDP partially coincide. In order to stabilize the output rate of the second layer, a method is proposed to reduce the spike correlation of the first layer of the network by increasing the number of input synapses to decrease the similarity of weights established by spike-timing-dependent plasticity in different first-layer neurons. Then, a two-layer network with the multiplication of input synapses is shown to be applicable to a toy classification task of Fisher’s Iris. The accuracy of classification is obtained to be 93 % by F1-score, which is at the level of the existing approaches for training spiking neural networks with rate encoding of the input.