This article discusses the use of universal adversarial as well as to improve the effectiveness of protection systems against robots and spam. In particular, the key features that need to be taken into account to ensure an optimal level of protection against robots and spam are considered. It is also discussed why modern methods of protection are ineffective, and how the use of universal adversarial attacks can help eliminate existing shortcomings. The purpose of this article is to propose new approaches and methods of protection that can improve the effectiveness and stability of protection systems against robots and spam.

Keywords: machine learning, clustering, data recognition, library Nanonets, library Tesseract

An original approach to describe airflow in the thin conic diffusor is suggested. It is based on approximate analytic solution of continuity equation. In addition simplified model of turbulence is combined. Reliability of derived formula is confirmed by comparison with finite-element solution for designed experimental setup. The elaboration is intended to direct computer simulation of multiphase flow.

Keywords: dust-air mixture, aspiration systems, turbulence, finite element modeling, separation diffuser, digital twin

A program for computer simulation of atomization of liquid fuels has been developed. A special calibrating experiment was carried out in a gravitational field. Verification of the computer model with experimental data is carried out, the correctness of simulation modeling is determined by the convergence of the results of static generalization. Tests of a model sample of the burner device were carried out, confirming the adequacy of computer simulation.

Keywords: burner, atomization of liquid fuels, mathematical modeling, fuel jet dispersion dynamics, nozzle

The article is devoted to the consideration of topical issues related to the study of the possibility of forecasting the dynamics of stock markets based on neural network models of machine learning. The prospects of applying the neural network approach to building investment forecasts are highlighted. To solve the problem of predicting the dynamics of changes in the value of securities, the problems of training a model on data presented in the form of time series are considered and an approach to the transformation of training data is considered. The method of recursive exclusion of features is described, which is used to identify the most significant parameters that affect price changes in the stock market. An experimental comparison of a number of neural networks was carried out in order to identify the most effective approach to solving the problem of forecasting market dynamics. As a separate example, the implementation of regression based on a radial-basis neural network was considered and an assessment of the quality of the model was presented.

Keywords: stock market, forecast, daily slice, shares, neural network, machine learning, activation function, radial basis function, cross-validation, time series

The problem of quality in construction and installation works at intermediate stages is being discussed. The role of technical supervision in the construction process is outlined. A proposed indicator, the "Quality Index," is presented for consideration, which reflects not only the quality of the works but also the effectiveness of the technical supervisor's work during the construction process.

Keywords: quality, construction quality, construction quality index, quality coefficient, regulations, technical supervisor engineer

In this paper, the possibility of applying graph neural networks (NN) to study the structure of copper centers of zeolites is considered. The dataset used for NN training was prepared using the FDMNES software based on the finite difference method and included more than 2100 Cu K-XANES spectra for Cu-MOR. The performed study demonstrated the capability of graph neural networks to reproduce the Cu K-XANES spectrum corresponding to a particular model of the copper center in the zeolite framework.

Keywords: zeolite, mordenite, atomic structure, XANES, machine learning, graph neural networks

The issues of preserving existing green spaces and other elements of the natural landscape during new construction are considered. Methods for preserving perennial plantings throughout the entire course of construction are proposed. If it is impossible to save the tree at the construction site, a method of transplanting it to another place is proposed. It is proposed at the stage of design and survey work to identify healthy trees that do not grow on the site allocated for a building (structure) under construction. Then solve the problem of locating the object on the ground in such a way as to preserve healthy perennial trees as much as possible. To do this, it is necessary to carry out the removal of the object to the area, moving it as far as possible from healthy trees. The distance required to protect the tree from external influences during work is equal to the projection of the crown on the ground plus 1.5 m. At this distance, it is recommended to make stationary fences for each tree. A tree transplantation scheme and a method for calculating its weight for the selection of equipment for digging and transportation have been developed.

Keywords: landscaping, construction, tree, tree transplanting, asphalting

This article discusses the forecasting of the collection of payments in post offices, taking into account seasonality and the use of machine learning. An algorithm for constructing a calculation model has been developed, which provides an opportunity for analysts of the Russian Post to make a monthly forecast of the collection of payments for each UFPS (Federal Postal Administration), taking into account seasonality. This model allows you to identify deviations from the norm in matters related to the collection of payments and more accurately adjust the increase in tariffs for services. The SSA algorithm is considered, which consists of 4 steps: embedding, singular decomposition, grouping, diagonal averaging. This information system is implemented in the form of a website using a framework ASP.NET Core and libraries for machine learning ML.NET . Then the forecast is evaluated using various methods.

Keywords: mathematical modeling, seasonally adjusted forecasting, collection of payments, machine learning, neural network

The article discusses the methods and approaches developed by the authors for the recommendation system, which are aimed at improving the quality of rehabilitation of the patient during respiratory training. To describe the training, we developed our own language for a specific subject area, as well as its grammar and syntax analyzer. Thanks to this language, it is possible to build a devereve describing a specific patient's training. Two main methods considered in the article are applied to the resulting tree: "A method for analyzing problem areas during training by patients" and "A method for fuzzy search of similar areas in training". With the help of these methods, it is proposed to analyze the problem areas of patients' training during rehabilitation and look for similar difficult areas of the patient to select similar exercises in order to maintain the level of diversity of tasks and involve the patient in the process.

Keywords: Recommendation system, learning management system, rehabilitation, medicine, respiratory training, marker system, domain-specific language, Levenshtein distance

The article deals with the issue of the reconstruction of the A-2 station in the context of the development of the North–South transport corridor. The relevance of the topic lies in the need to master the growing volumes of transportation from Russia to the countries of Southeast Asia and back in connection with the current reorientation of the main export cargo flow and economic ties of Russia. The analysis of the existing volumes of work of the station is carried out. To develop proposals, an analysis of the volume of work of the station was carried out, according to which there was a decline in production over the previous period. Measures are proposed for the effective development of the expected cargo flow. Statistical modeling methods, queuing systems theory and feasibility studies were used as methods. As a result, the proposals made are aimed at implementing the effective operation of the A-2 station. The reconstruction of the station makes it possible to ensure the development of increased cargo volumes and thereby receive additional income.

Keywords: relevance, transport corridor, station, goal, analysis, cargo turnover, volume of work, forecast, activities, expected effect

The problem of fake diplomas of education causes alarm and concern to society. In the digital age, falsification has reached great proportions. In this regard, a mechanism for recording and confirming the authenticity of diplomas using technology is proposed. A sector-token method of accessing a blockchain record is proposed. The recording technology and the blockchain formation model are shown. The proposed technology guarantees that the diplomas are genuine, protected from forgery, belong to the specialists who received them.

Keywords: blockchain, data protection, diploma forgery, educational institution, authentication

The possibilities of a little-studied method for obtaining nanosized materials of electronic engineering with a given substructure, the zone sublimation epitaxy (ZSE) method, are discussed. In the work, it is combined with the method of gradient liquid phase epitaxy (GLE). A specific feature is mass transfer in a two-phase medium (a solid substrate and an inert gas phase acting as a transport medium) with preliminary deposition of a matrix layer formed from the melt. A feature of the sublimation process in the study was the crystallization of low-melting iron-silicon eutectic. A mathematical model of the process was proposed and compared with the experimental results. Island structures of the composition silicon (more than 90%), iron (up to 8%) and chromium (about 1.5%) have been obtained. Their parameters and size distribution were studied. A Solver-HV scanning probe microscope and a Quanta-200 scanning electron microscope were used. The study shows that the use of sublimation transfer transients makes it possible to reproducibly form doped silicon nanolayers and transform them into regular mesostructures.

Keywords: microsize growth cell method, zone sublimation epitaxy, gradient liquid phase epitaxy, island nanostructures

A structural model of an intelligent classifier of unstructured textual data according to the degree of confidentiality is presented, which is a two-level cascading ensemble of classifier models. The meta-model of a fully connected neural network architecture, which has the greatest impact on the classification efficiency, is highlighted. The multi-criteria task of configuring the intelligent classifier is decomposed into the task of selecting configurable hyperparameters of the meta-model and the task of selecting their values. Taking into account the selected hyperparameters of the neural network meta-model, the multi-criteria task of selecting hyperparameters to be configured is presented in the form of a hierarchy that includes the goal, criteria and alternatives. A method for selecting configurable hyperparameters of an intelligent classifier of unstructured text data by the degree of confidentiality based on the hierarchy analysis method has been developed.

Keywords: DLP system, unstructured text data, intelligent classifier, hyperparameters, hierarchy analysis method

The article describes the algorithmic realisation of a software module for evidence of learner’s identification in the testing process. The advantage of this module is simple operation, ease of implementation and execution as well as convenience of application by various categories of users. The need for such a module was engendered by the problem of examinee identification during testing in the e-learning system. The technology of program module operation is based on forming questions with the use of information stored in the learner’s personal account; the operation result is demonstrated through confirmation or non-confirmation of the examinee’s identity in real time.

Keywords: distance learning, identification, student testing, software module, learner’s personal account

In order to provide information support for decision-making on the issuance of bank guarantees for the execution of a contract in the field of public procurement, it is important for banks to obtain historically accumulated information on the execution of government contracts. This is necessary to assess the possibility of the supplier's performance of his future contract. This can be done by collecting and aggregating information about contracts from the Unified Information System in the field of procurement. The paper proposes to use IT technologies and data analysis to predict the performance of the contract and identify bona fide suppliers. In the work, a selection of primary data on contracts was formed for modeling using the parsing of the FTP server of the Unified Information System in the field of procurement, and the parsed data was preprocessed for use in machine learning models.

Keywords: information system, data analysis, government contract, data parsing, machine learning