Asynchronous Hierarchical Federated Learning: Enhancing Communication Efficiency and Scalability

This study introduces Asynchronous Hierarchical Federated Learning (FedAH), a novel framework designed to tackle communication bottlenecks and scalability challenges in federated learning (FL) systems. By integrating asynchronous updates with a hierarchical communication structure, FedAH reduces the computational load on the central server, accelerates convergence, and addresses client heterogeneity. Experiments conducted on the CIFAR-10 dataset using a Convolutional Neural Network (CNN) demonstrate that FedAH achieves faster convergence and significantly lowers communication overhead compared to synchronous methods. However, it also introduces challenges related to system stability, highlighting the need for optimization of learning rates and hyperparameters. Future work will focus on enhancing system stability and scalability through refined hyperparameter tuning and expanding evaluations to diverse datasets and architectures to validate the framework’s generalizability. FedAH represents a promising advancement in efficient and scalable federated learning systems, paving the way for broader applicability in large-scale distributed networks.