Differential Models for Representation and Optimization of Architecture and State Modes of AI Networks

The evolution of the mass use of artificial intelligence is analyzed, and the direction of research related to the optimization of the architecture and state modes of AI networks is substantiated. It is shown that the performance of AI networks largely depends on their topological characteristics. Their representation in the form of an unweighted and undirected graph is proposed, and a number of mathematical models for topology optimization are developed. The application of the minimax strategy for optimizing the modes of operation of AI networks in the case of the worst combination of the intensity of the transition action flow with an arbitrary counteraction flow law is substantiated. Based on the theory of differential transformations, differential mathematical models are proposed for optimizing the dynamics of the probabilities of the states of AI systems in time. To ensure the necessary productivity of information exchange in AI systems and take into account the conditional balance of traffic volume and sufficient bandwidth at all levels of the hierarchy, a new approach is proposed, which is based on methods for reducing topological network distances as the basis for reducing transit traffic