FedOPT: federated learning-based heterogeneous resource recommendation and optimization for edge computing

Resource recommendation in edge computing relies on distributed resource alignment across multiple servers and interconnected networks. Consequently, addressing issues, such as resource overloading and resource planning, becomes paramount in research endeavours. Federated learning emerges as a dynamic method for processing distributed data within a heterogeneous networking infrastructure. The interconnectedness of nodal source nodes has escalated due to the significant data volume generated and the management challenges associated with resources. This article introduces an innovative approach known as FedOPT, designed for heterogeneous resource recommendation and pooling within an optimised framework. The FedOPT technique selectively extracts pertinent data from interconnected and layered nodes in operational spaces. It incorporates a gradient descent parameter to evaluate resource utilisation on each edge device. FedOPT analyzes the convergence bounds of distributed gradient parameters in conjunction with a resource recommendation and coordination algorithm. The recommendation technique's performance is assessed in real-time node communication and data coordination scenarios to ensure the dependable prediction of node paths and resource allocations within dynamic networking environments. The experimental setup and results indicate significant optimizations, resulting in an impressive accuracy rate of 96.43% in resource recommendations. This accuracy extends across various machine learning models and distributed networking infrastructure models.