Current level of development in the sphere of deep learning allows replacing existing domain-specific algorithms for military simulation with approximating neural networks. Hyperparameter search allows finding network’s architecture, appropriate for a task. This work describes that process for the task of pre- dicting area of optical visibility, taking a fragment of a digital map as input and proposes ancillary architectural solutions for stitching building blocks together, assuring their conformation for performing search among their pos- sible combinations within the architectural space. The final proposed result is a channel-wise attention U-Net with an encoder, based on ResNet50 backbone.

The aim of this study is to develop and validate a model for predicting injury risk in athletes using machine learning methods to analyze time-series physiological data. The research focuses on data from runners with the high qualification, including heart rate (HR), heart rate variability (HRV), and training load, allowing for the assessment of an athlete’s physiological state over time and the identification of potential health risks.

The proposed prediction model, based on a Long Short-Term Memory (LSTM) recurrent neural network, enables the identification of periods of increased injury risk in athletes and allows for adjustments to the training process to prevent injuries. Experimental results on real-world data demonstrated a high prediction accuracy of 85%, confirming the effectivenes of the proposed approach in forecasting injury probability. Despite the model’s success, the authors recognize the need for further longitudinal studies with an expanded dataset to ensure a more precise validation of the proposed method.

The functional components of a multi-agent intelligent system have different physical or logical structures and provide processing of task flows with different intensities. From the point of view of the effectiveness of the tasks set, obtaining a consolidated result and achieving a common goal by the agents, this system is considered as a single object, an integrated entity. At the same time, the overall efficiency of its components is assessed by a common parameter by which it can be compared with other architectural variants of multi-agent systems. In this regard, it is proposed to evaluate the effectiveness of the multi–agent system by a conditional extremum – the total number of tasks in the queues of all agents of the system, provided that the ability to provide the necessary margin for the load factor of each agent limits the zone of its stable functioning. It is shown that a random search algorithm can be used to optimize the task processing process, which consists in randomly selecting points in the space of possible solutions, evaluating their quality using an objective function, and preserving the best of the solutions found. The considered task of minimizing the objective function – the total number of tasks in the queues of all agents of the system is interpreted as performing approximate nonlinear optimization using the Lagrange multiplier method. As an example of the implementation of the proposed method for optimizing the task processing process of a multi-agent system, the results of a computer experiment to determine the minimum value of the objective function are given. Based on the specified solution to the optimization problem, a technological algorithm for the functioning of a task distributor in a multi-agent intelligent system is proposed.

A robot movement modeling algorithm with obstacle avoidance using the Q-learning machine learning method is proposed. Q-learning allows for preserving the rewards obtained during modeling by performing optimal actions in each specific state. The Q-table contains information about the state and actions of the robot. Storing the Q-table in the blockchain using IPFS (InterPlanetary File System) technology ensures reliable and decentralized storage of data about the robot's states and actions. Content addressing in IPFS separates the data from its location and retrieves files from multiple sources in a peer-to-peer mode. A computational experiment for the proposed algorithm was conducted using a robot movement simulation environment. In the Gazebo 11 visualization package, it was shown that using the new algorithm, obstacles are avoided faster (by 59.8 %) compared to the previous version of the algorithm.

The article provides an analysis of existing methods for reducing the likelihood of blocking the movement of public transport vehicles at controlled intersections. On the section of the street and road network in the area of the intersection of Nezavisimosti Avenue – Masherov Avenue – Kozlov Street, flexible control algorithms have been developed to reduce blockages of the movement of route passenger vehicles by other vehicles based on intelligent transport systems, and solutions have been proposed for the use of modern means of traffic light control. Using simulation micro-modeling, an assessment of the effectiveness of the proposed solutions and their efficiency was performed.

This study analyzes the distribution of real objects and synthetically augmented classes, as well as their impact on machine learning models. The training results of logistic regression, decision trees, random forest, and SVM models on synthetic data were compared with those obtained on a dataset of real objects. Experimental results showe  that the use of synthetically augmented data improves the accuracy of classification models, with particularly noticeable improvements observed in some algorithms.

The vulnerability of the technology for generating a common cryptographic key using synchronized artificial neural networks (ANN) is considered in a relatively new type of attack called two-way. The essence of the attack is that a cryptanalyst, listening to an open communication channel, creates two identical artificial neural networks, one of which he synchronizes with the network of one of the legal subscribers, and the second with the network of another legal subscriber. By comparing the vectors of weight coefficients of attacking networks, the cryptanalyst is able to determine the moment of full synchronization of the networks of legal subscribers and the value of the generated secret number. Next, we examine the possibilities of a two-way attack using various models of secret number generation.

The article is devoted to the study of the features of using decision support systems in creating automated intelligent systems for managing educational processes. The substantiation of the principles and methods of synthesizing an intelligent system, the features of using various mathematical models of regulators, their advantages and disadvantages is carried out. A reasonable conclusion is made that the use of traditional approaches based on the use of theoretical provisions of the classical theory of automatic control as applied to the synthesis of control systems for social systems, which include educational process control systems, does not allow achieving the desired result. It is proposed to use the principles and methods of building intelligent systems when building an educational process control system, the main difference of which is that the system itself synthesizes the control goal based on external conditions, the knowledge base available in the system and motivation for the tasks of the system functioning. The construction of a decision support system as an element of the control block is considered, the main element of which is a person - a manager (teacher), who directly manages the object, which is also a person - a student (pupil, student, listener). A structural diagram of the construction of an intelligent system for managing information flows of the educational process and a model of influence on the control object are proposed. The basis of the control system is a dynamic expert system that carries out predictive modeling of the results of control action formed on the basis of the theory of fuzzy sets. An example of modeling information flow control is given, demonstrating a qualitative picture of the processes occurring in the system taking into account external influences on the control object.

The article proposes an approach to solving the problem of optimal distribution of enrolled applicants by study groups, taking into account the level of motivation of those being distributed. The results of this work can be used by universities as a natural administrative tool to achieve better educational results in the context of academic heterogeneity.