A knowledge-driven model selection and resource management method with information entropy

The rapid advancement of the Internet of Things (IoT), mobile edge computing (MEC), and artificial intelligence (AI) technologies is accelerating the emergence of novel services and applications with substantial computational demands. This phenomenon has resulted in an exponential growth in the complexity of devices and networks, which has engendered the demand for the evolution of network resource management paradigms. Traditional resource management strategies focus mainly on reducing device delay and energy consumption, but relatively lack attention to the selection and sharing of intelligent models among multiple devices. To address these challenges, this paper proposes an information plane (IP) framework for evaluating the representation capability of models to support complex network resource management services. Furthermore, we design a deep reinforcement learning (DRL)-based resource management scheme, which considers delay and energy consumption and takes the model representation capability as a new indicator, thereby achieving a knowledge-driven intelligent model selection and resource management. Specifically, we first use the IP framework and blockchain technology to introduce model evaluation metrics and identify the best-performing model in the network. Then, we design the resource management problem as a Markov decision problem (MDP) to achieve the optimal decision and resource allocation of the system. Experimental results show that compared to traditional resource management schemes in MEC-assisted IoT scenarios, the proposed scheme can effectively select a suitable intelligent model for IoT devices and optimize the cost of the system, and its performance is better than other schemes.Specifically, compared with traditional schemes, the proposed scheme achieves up to a 7% improvement in model representation capability, while reducing delay by 25% and energy consumption by as much as 65%.