Resource-Efficient and Delay-Aware Federated Learning Design under Edge Heterogeneity

• DOI: 10.48550/arxiv.2112.13926
• Publication Date: 2021-01-01
• Authors: David Nickel; Frank Po-Chen Lin; Seyyedali Hossein...; Nicolò Michelusi; Christopher G. Br...
• Keywords: Edge device

Federated learning (FL) has emerged as a popular technique for distributing machine across wireless edge devices. We examine FL under two salient properties of contemporary networks: device-server communication delays and device computation heterogeneity. Our proposed StoFedDelAv algorithm incorporates local-global model combiner into the synchronization step. theoretically characterize convergence behavior obtain optimal weights, which consider global delay expected local gradient error at each device. then formulate network-aware optimization problem tunes minibatch sizes devices to jointly minimize energy consumption training loss, solve non-convex through series convex approximations. simulations reveal that outperforms current art in FL, evidenced by obtained improvements objective.