A Federated Meta-Learning Aided Intelligent Edge Framework by Using the Parameter Optimization Approach

Edge intelligence can enable fast intelligent services by integrating edge computing with machine learning, thereby facilitating intelligent information processing for Internet of Things (IoT) devices on the edge. However, intelligent data processing at the edge may expose IoT devices to the risk of private information leakage. To mitigate this issue, we propose a federal meta-learning-aided data processing framework to cope with complex tasks in edge IoT networks. Unfortunately, communications between edge IoT devices and edge servers in federated frameworks incur significant overhead. To address this challenge, we propose a parameter optimization algorithm that alleviates communication costs between edge IoT devices and edge servers, thereby reducing classification errors induced by parameter optimization. Moreover, the convergence of the federated meta-learning method is derived, which theoretically confirms the feasibility of the proposed approach. Simulation results demonstrate that the error minimization-based quantization compression optimization algorithm can substantially enhance communication efficiency while incurring only negligible precision losses.