Vol 31, No 2 (2025)

The issues of process specification of the model of a specialized subscriber protocol of an automated system for the exchange of documentary information using Petri network notations and time logic are considered. The semantics of document exchange is determined by the paradigm of alternation of quasi-parallel input-output processes of a reacting system. The formalization of a number of properties of the semantics of an automated system in abstractions of temporal logic is shown.

The paper proposes a method for implementing the translation of logical addresses of host requests of stored data into physical addresses of data located in an array of non-volatile NAND Flash memory. The main limitations of working with NAND Flash memory are described, and a set of mechanisms for solving the problem of address translation is proposed. Additionally, optimization methods of the translation table cells are described to achieve high optimization of interaction with the memory buffer allocated for storing the address translation table. The proposed methods and the described features can be used in the development of address translation units of modern high-performance solid-state drive (SSD) controllers with high logical capacity.

In recent years, cyberattacks, including DDoS attacks, on the critical information infrastructure of the Russian Federation have resulted in financial losses for companies, enterprises, individuals, universities and even hospitals. The damage reaches trillions of roubles,and on average, each large online shop that has been attacked can lose up to 600,000 roubles a day. And this is despite the fact that most companies have their own equipment and software to detect and filter DDoS attacks, or use the services of providers/data centres.

The main reason is that not all companies, providers and data centres have sufficient capacity to filter DDoS attacks of various types and types. In addition, an equally important reason is the misconfiguration of physical servers and network equipment ranging from switches to software-defined networks (SDNs)/content delivery networks (CDNs).

Thus, given the importance and necessity of ensuring the availability of critical information infrastructure in the era of digital economy, this paper presents a comprehensive systematic review of DDoS attack types and their intelligent filtering techniques.

The main findings and results of this study open up the possibility of implementing next-generation systems based on neural networks and computational clusters to analyse network traffic and detect DDoS attacks. In addition, these systems will help to solve existing critical problems, the main ones being the speed of response to emerging cyberattacks and the quality of filtering unauthorised network traffic.

The methods for solving the problems of image inpainting and image super-resolution by means of image generation with neural networks are considered. Generative and adversarial neural networks are created and trained to solve them. It is shown that, in a wide range, the recovery quality almost does not depend on the fraction of damaged pixels, that adding residual blocks does not lead to its improvement, and that the generative adversarial network for resolution increase gives better results than the bicubic interpolation.

A gradient descent algorithm for optimal control of dynamic systems is developed taking into account the free right end of the trajectory and control constraints. A feature of the algorithm is the possibility of its generalization for various boundary conditions. The main attention is paid to the mathematical justification of the method and its application to problems of chemical kinetics. Numerical experiments are conducted confirming the efficiency of the algorithm for optimizing real chemical processes.

The restoration process of MRI images of a patient’s abdominal region is considered based on multi-task learning, including the creation of super-resolution images and noise reduction using deep learning methods. An improved RIRGAN model is proposed by adding a noise reduction module that compensates for additive noise and nonlinear noise. The proposed multi-task model called MT-RIRGAN is trained using a complex loss function consisting of pixel loss, perceptual loss, adversarial loss, and total variation loss. Experiments demonstrate good recovery results of MRI images while preserving the original visual structures important from the point of view of medical diagnostics.