Roman Meshcheryakov

This study presents a transformer-based Generative Adversarial Network framework designed to model and replicate complex behavioral patterns associated with account manipulation in cybersecurity environments. The primary objective is to generate realistic synthetic user behaviors that emulate malicious activities while ensuring data privacy, thereby providing an ethical and controlled foundation for developing Artificial Intelligence (AI) driven defense mechanisms. The proposed architecture integrates progressive adversarial training, feature matching, and hybrid normalization to stabilize learning and enhance fidelity. A Transformer encoder captures long-range temporal dependencies in behavioral data, while a multi-scale attention discriminator identifies manipulative patterns across varying time scales. A noise-adaptive pre-processing pipeline and composite authenticity evaluation protocol enhance the alignment of synthetic sequences with authentic behavioral statistics. The model surpasses recurrent neural baselines in experimental evaluation, attaining a Fréchet Inception Distance of 18.3, precision of 0.82, and recall of 0.76, alongside enhanced training stability (37 %) and improved behavioral coverage (29 %). Using t-distributed Stochastic Neighbor Embedding and Mahalanobis distance metrics to visualize the data shows that 89 % of the behaviors that were real and those that were generated were the same. This means that the mimicry was very accurate and the timing was very consistent. This study establishes a resilient AI framework for behavioral simulation, facilitating the advancement of adversarially trained, adaptive cybersecurity systems. It stresses even more the need for ethical governance, responsible AI use, and behavioral biometrics to reduce the dual-use risks that come with generative technologies in modern cyber defense.

   Purpose of research. Methods for embedding digital watermarks into digital objects constitute a scientific field at the intersection of information security and digital signal processing. Digital watermarks serve to ensure the authentication of digital objects by being inseparably embedded into them through the manipulation of data elements that constitute the digital object.    A key requirement  for digital watermarking methods is robustness, which entails that a watermark remains intact even after transmission over a heavily noisy channel. Therefore, the purpose of the research is to evaluate the robustness of watermarks to noise arising from the implementation of neural network removal attacks using the example of a watermark embedding algorithm that is robust to classical laboratory attacks.    Methods.  The study was conducted using the example of an algorithm for embedding a digital watermark into the coefficients of the discrete cosine transform of digital images using metaheuristic optimization. To implement a BlackBox removal attack, three neural network models were trained: one based on a convolutional neural network, a recurrent neural network, and a multilayer perceptron.    Results. According to the results of experiments aimed at removing watermarks from the watermarked images as part of the BlackBox attack, the convolutional neural network architecture demonstrated the greatest efficiency. The watermarking algorithm demonstrated resistance to watermark removal by two other neural network models.    Conclusion. The conducted research showed that the resilience of a digital watermark embedding algorithm to classical attacks does not guarantee resilience to neural network-based removal attacks. Resilience to this class of attacks must be incorporated at the stage of embedding algorithm synthesis. This can be achieved by reducing the determinism in the selection of watermark embedding areas in the image and by implementing anti-attack techniques at the embedding optimization stage.

Purpose of the article: the design of a multi-level monitoring and response system for impacts, ensuring the resilience of complex ergatic systems through the use of a backup control loop and adaptive resource allocation. Research methods: system analysis, modeling, architecture synthesis, resource allocation. Research results: a multi-level architecture of a monitoring and response system for impacts has been developed, aimed at improving the resilience of complex ergatic systems. A structural solution is proposed, including primary and backup control loops, which ensures continuous monitoring and coordinated response even in cases of partial degradation or failure of the communication infrastructure. A mechanism for adaptive resource reallocation between architecture components has been designed, ensuring efficient system operation under variable loads and limited computing and network capabilities. The theoretical significance of the work lies in advancing scientific knowledge on the design of multi-level architectures for the protection of ergatic systems, ensuring their functionality under complex impacts. The practical significance is determined by the possibility of applying the designed architectural solutions in the creation and modernization of distributed automated complexes of various purposes to increase their resilience and the efficiency of monitoring processes. Scientific novelty: for the first time, a multi-level architecture of a monitoring and response system for impacts is proposed, implementing a separation into primary and backup control loops to ensure operational resilience under communication degradation and failures. For the first time, a mechanism for adaptive resource reallocation between system levels has been developed, allowing the maintenance of monitoring and response efficiency under variable loads and limited computing and network capabilities.

Advances in neural networks have enabled unmanned aerial vehicles (UAVs) to detect and recognize objects in real time, which has facilitated the use of UAVs autonomously in a variety of scenarios, including fire detection in emergency situations. The paper reviews a number of existing neural network-based detection algorithms, including convolutional neural networks, regional convolutional neural networks and their variants, deep neural networks with convolutional long short-term memory (ConvLSTM), methods integrating deep learning with correlation filtering through self-training, Siamese neural networks for target tracking, and the YOLO (You Only Look Once) family of algorithms. The main characteristics and differences between neural network algorithms are described, and a comparison of their performance in terms of mean average precision (mAP) and frame rate per second (FPS) is given. The conclusions of the article provide insight into the trade-offs between accuracy, speed and task-specific requirements in detection tasks, which allows one to make an informed choice on the use of one or another algorithm.