Information and Communication Technologies in Education, Research, and Industrial Applications

The paper is devoted to extending the coalgebraic technique of analysing Discrete Systems to the class of Random Systems with Output. To achieve this goal, the authors propose a randomization procedure, which uses the endofunctor of finite distributions as a tool for constructing the Random System based on a Deterministic System. Given the particular importance of final coalgebras for the analysis of systems using coalgebraic techniques, a system that claims this role and a family of morphisms, which in this case is a family of anamorphisms, are constructed. All formal constructions are illustrated by examples for the convenience of readers who do not fully master the language of category theory and coalgebraic techniques. Establishing the fact that the constructed artefacts describe the final Random System with Output and the corresponding family of anamorphisms remains an open problem.

This paper presents a software system called “Algebraic Virtual Machine (AVM), which handles industrial hardware specifications, programs in different languages, and models in algebraic language. It uses the formal algebraic methods that were developed in the scope of behavior algebra and help to resolve the problems of verification, analysis, testing, and cybersecurity. The new version of AVM will include the possibilities to formalize continuous process and significantly extends the usage of formal methods. It permits the possibility of creating your own methods and theories and trying them with industrial examples with minimal efforts. The machine learning technique is used for the definition of formal method efficiency, and the classification model is trained during algebraic processing. The formalization and checking for resistance of blockchain attack is considered as case study.

The chapter is devoted to the actual scientific and technical problem of implementation the secure remote control of overlay networks, presented in the form of intelligent mobile objects. The specificity of this class of networks is the use of existing cellular communication networks as a data transmission medium. The security of data transfer processes between the nodes of the overlay network is ensured by using the technology of private virtual networks. The chapter also shows options for using their modifications: with route regeneration for high-mobility networks and technology for nesting tunnels in low-mobility networks. Therefore, additional tasks were solved related to the choice of the trajectory of the nodes behavior in autonomous operation, ensuring a given level of survivability and reliability of the overlay network in terms of routing dynamics. The chapter shows the results of modeling various network attacks aimed at intercepting and replacing traffic in the chain “intelligent mobile object - control node”. A mathematical model is proposed for regenerating data transmission routes in a network from the point of view of Markov and semi-Markov processes. Examples of solving the problem of finding a quasi-optimal solution under conditions of lack of time for the regeneration of data transmission routes are given using the Wald scheme as an example. The obtained numerical studies of the proposed models prove the efficiency of the proposed approaches. Using the proposed approaches, it is possible to create full-fledged network segments within the global urban networks that provide the concept of “smart city”. Numerical studies have shown that data transmission with a sufficient degree of reliability is achieved while maintaining losses in the range from 7% to 45%. Otherwise, IMOs operate offline. The proposed models can be used in the development of methods for reengineering of existed and creating of modern resilient networks for IMO based on idea of proactive dynamical reconfiguration and real-time evolution considering changing requirements and environment parameters.

Research into generative representations of complex data is a rapidly expanding field in machine learning. In this work we propose and evaluate a process of production and analysis of informative low-dimensional latent representations of real-world images with neural network models of unsupervised generative learning. A model of convolutional autoencoder based on VGG-16 architecture was used to produce low-dimensional generative representations of two datasets of aerial images and the characteristics of distributions of several classes of images of terrain studied. An analysis of latent distributions of terrain classes demonstrated a landscape of compact density structures for most studied classes with good separation of concept regions. The results of this work can be used in developing methods of effective learning in problems and environments with a deficit of labeled data based on the concept-sensitive structure in the latent representations that emerges in the process of unsupervised generative self-learning.

Discourse connectives are words and phrases that mark discourse structure by expressing coherence relations explicitly. In this paper, we present the first complete lexicon of Ukrainian discourse connectives, fully annotated with syntactic and semantic properties. We compare the inventory of connectives in Ukrainian with several other European languages.

We demonstrate the usefulness of the lexicon in two case studies. First, we carry out a manual discourse connective annotation of written and spoken Ukrainian text, which leads to the discovery of additional connectives, and enables us to compare the discourse structure of spoken vs. written language in Ukrainian. Second, we automatically identify explicit connectives using a rule-based system, by disambiguating connective candidates from the lexicon. The identification of explicit discourse connectives is the first step of automatic discourse parsing. On a cross-domain evaluation of the automatic connective detection, we reach an \(F_1\)-score of 0.69.

One of the crucial problems of natural language processing for languages such as Ukrainian is lack of datasets both unlabeled (for pretraining of word embeddings or large deep learning models) and labeled (for benchmarking existing approaches).

In this paper we describe a framework for simple classification dataset creation with minimal labeling effort. We create a dataset for Ukrainian news classification and compare several pretrained models for Ukrainian language in different training settings.

We show that ukr-RoBERTa, ukr-ELECTRA and XLM-R large tend to show the highest performance, although XLM-R large and ukr-ELECTRA tend to perform better on longer texts, while ukr-RoBERTa substantially outperforms other models on shorter sequences. To analyze these results deeper, we use SHAP to gain insights into the patterns these models learnt.

We publish this dataset on Kaggle (https://​www.​kaggle.​com/​c/​ukrainian-news-classification/​) and suggest to use it for further comparison of approaches for Ukrainian text classification.

This paper considers issues of business process model structuredness, which are mostly related to inaccurate usage of gateways. According to related work in a business process model structuredness domain, split gateways ought to match respective join gateways of the same type, while the existing mismatch measure allows evaluating model structuredness only by degrees of split and join gateways. Thus, the current measure of process model structuredness is not accurate enough and business process model shortcomings could remain undetected, which negatively impacts model understandability, maintainability, and may increase the error probability of business process models. Therefore, error fixing effort and related expenses may grow exponentially during later stages of the information system lifecycle. Hence, we have proposed an improved gateway mismatch measure, a model to detect business process modeling errors and formulate recommendations to achieve sound models, and a model for effort estimation of business process modeling error fixing. An information technology for detection and fixing effort estimation of business process modeling errors was designed and implemented to perform experiments with a large set of business process models that belong to different industries. Sample business process models that were used for calculations are demonstrated, as well as obtained results are analyzed and discussed. Conclusion and future work are formulated.

We researched how to use financial technology in the finance industry on the example of robo-advisors; defined the basic functionality of a robo-advisor; got the robo-advisors implementation based on analysis of the most popular financial services. We compared their functions, composed a list of critical features and described the high-level architectural design of a general robo-advisor tool, scope of using robo-advisors, their key features, and a brief overview of existing solutions. Using Markowitz model, we set up a concept of using a robo-advisor by investors who have different attitudes towards risks. Our goal is to cover the main features of financial robo-advisor and to describe a high-level architecture for such applications using prediction of financial instruments rates to rebalance investment portfolio. We have defined the main modules that represent the architecture of a typical robo-advisor. We also described different techniques, which could be applied building a personalized investment portfolio. We considered ARIMA models to predict stock prices. The experimental part demonstrates how to use LSTM neural networks and multiple linear regression techniques in the scope of the Robo-Advisor profitability-forecasting module.

A brief review of molecular modeling methods and specialized software and their use for creating and researching molecular models was considered. The authors considered the algebraic approach to modeling molecular interactions in some environments to determine the triggering of the studied properties. In particular, the article describes the results of the first steps of building a tool for the study of molecular, and in particular, biomolecular interaction, based on the formalism of behavioral algebra and insertion modeling. The experiment’s results of applying the proposed approach to modeling atoms interaction (creating of atomic bonds - valence method), constructing the electronic configuration of the molecule/substance (molecular orbitals method), and calculating their main parameters are given. The formalization and properties analysis is considered using the insertion modeling platform.

The performance of the Raspberry Pi 4B computer was evaluated for three cases. First, a Raspberry Pi heterogeneous MPICH cluster with weight-based load balancing and fuzzy estimation of node computational performance was designed. Fuzzification, formation of fuzzy rules, fuzzy inference, and defuzzification were employed to determine the performance weights. In the cluster with two Raspberry Pi 4B boards with 2 GB RAM and Raspberry Pi 64-bit OS and one Raspberry Pi 3B board with 1 GB RAM and Raspberry Pi 32-bit OS, the recommended performance weights are (5, 5, 1), respectively. The developed Python program for the prime numbers finding algorithm employs the proposed weight-based load balancing, which is approximately five times faster than the basic algorithm with equal loading for the maximum integer of 300000. Second, the MPICH cluster with two nodes in two virtual machines located on two different Raspberry Pi 4B boards with Ubuntu Server for ARM on the hypervisor VMware ESXi ARM Fling shows the mean signed deviation −34.01 s regarding the Raspberry Pi 64-bit OS for the maximum integer of 300000. Third, the performance of the Raspberry Pi 4B 8 GB computer with Windows 11 ARM OS was compared with the laptop Lenovo G510 with Intel Core i7-4700MQ and Windows 10 64-bit OS using the combinatorial optimization algorithm implemented in the 32-bit Windows app. The Raspberry Pi 4B 8 GB consumed approximately six times more power. Thus, the Raspberry Pi 4B single-board computer is recommended for executing low-performance applications and/or short-term processing of high-performance tasks.

The results of experimental testing of the developed software for matching the focus of the student’s gaze with the structure of the training content on a computer monitor are presented in this paper. The use of widespread equipment is assumed: a laptop with a built-in camera or one additional camera. Initial processing of the face image, selection of eye areas is carried out using the OpenCV library. An appropriate algorithm for calculating the center of the eye pupil and the point on the monitor corresponding to the current focus of the gaze has been developed. The influence of the system calibration process with different schemes of calibration point display, its delay time on the screen and location of the additional camera according to the accuracy of the calculation of the coordinates of the gaze focus is investigated. Based on the performed experiments, it was defined that the error of gaze focus recognition with using two cameras can be reduced to 4–10%. The experiment in order to improve the calibration processes and evaluate the capabilities of the developed software for use on a laptop with only one built-in camera involving a group of students was carried out. The proposed approach makes it possible for objective measurement of each student working time with one or another part of the content. The lecturer will have the opportunity to improve the content by highlighting significant parts that receive little attention and simplifying those elements that students process for an unreasonably big amount of time. It is planned to integrate the developed software into the LMS Moodle in the future.

In this paper the algorithm for adaptive testing of students’ knowledge in distance learning and an assessment of its effectiveness in the educational process has been proposed. The results of the study are based on the achievements of modern testing theory IRT. The aim of the study was to build an adaptive testing algorithm that allows you to objectively assess students’ knowledge and assess the quality of test items. To achieve this goal, a mathematical model of modern testing theory IRT, namely the Rasch model, was used. The effectiveness of the proposed algorithm for the objective assessment of students’ knowledge has been experimentally shown. The correspondence of the empirical data to the Rasch model was assessed based on the Pearson’s chi-squared test. The quality of test items was provided based on the Rasch model.