Нейронна мережа з фізичною інформацією для оброблення сигналів та зв’язку мережевого трафіку

• Дата: 2026
• Автори: Zolotukhin, Oleg, Kudryavtseva, Maryna, Bodyanskiy, Yevgeniy, Filatov, Valentin, Antilikatorov, Andriy, Kalinin, Dеnys
• Формат: Стаття
• Мова: Англійська
• Опубліковано: The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute" 2026
• Теми: оцінка адаптивного моменту; штучний інтелект; штучна нейронна мережа; візуалізація даних; динамічний нейрон; градієнтне зниження; інформоване машинне навчання; алгоритм навчання; використання пропускної здатності мережі; оптимізація мережевого трафіку; нейро-нечітка логіка; Python
• Онлайн доступ: https://journal.iasa.kpi.ua/article/view/358086
• Назва журналу: System research and information technologies

Problem of signal processing and network traffic optimization is solved at the hardware level and is interesting, modern and relevant from the point of view of the application level. It is necessary to propose an approach that combines machine learning methods with network bandwidth tasks and traffic over the network. To solve this problem, it is proposed to use concept of Informed Machine Learning (IML), that is the Taxonomy of IML, the principles of constructing deep machine learning systems based on information about the physical properties of the data transmission network under study. The platform for developing is deep machine learning models using PINN neural networks. The PINN represents the class of deep learning algorithms that can integrate data with or without physical processes description. As an algorithm it is proposed to use the popular algorithm in the field of deep learning – ADAM (Adaptive Moment Estimation) for optimizing network traffic. Using the PINN trained with the ADAM algorithm to transmit data the efficiency has increased. Thanks to this method, it is possible to obtain a low noise signal in practice, due to which network traffic is optimized.