Collaborative Optimization for Resource-constrained Federated Learning in Large-scale IoT Networks
Haihui Xie, Shuwu Chen, Teng Sun, Junhui Zhao, Minghua Xia · 2024
Large-scale Internet-of- Things (IoT) networks enable intelligent applications and services, such as autonomous deriving. As many users generate various datasets, federated learning in distributed IoT networks emerges from learning from distinct datasets. To realize efficient and reliable communications in distributed networks, we propose a collaborative optimization model for resource-constrained federated learning using a joint design of wireless resource allocation and expected learning losses. Precisely, we start to formulate a learning-oriented power allocation problem. Then, we derive a convergence bound and build the relationship between communications and learning. At last, we perform an optimal algorithm based on majorization-minimization frameworks. Thanks to the high parallelization of the proposed algorithm, extensive experimental results corroborate that optimal power allocation in distributed networks benefits efficient federated learning compared to the state-of-the-art benchmark algorithms.