Adaptive Federated Learning in Heterogeneous Wireless Networks with Independent Sampling

TL;DR

This paper introduces an adaptive independent client sampling strategy for federated learning in heterogeneous wireless networks, optimizing wall-clock training time by considering data and system heterogeneity, and demonstrates its superior performance through real-world experiments.

Contribution

It proposes a novel independent client sampling method tailored for heterogeneous wireless environments, with a convergence analysis and an adaptive bandwidth allocation scheme.

Findings

Significant reduction in training time compared to existing sampling methods.

Effective handling of data and system heterogeneity in federated learning.

Validated improvements through real-world wireless network experiments.

Abstract

Federated Learning (FL) algorithms commonly sample a random subset of clients to address the straggler issue and improve communication efficiency. While recent works have proposed various client sampling methods, they have limitations in joint system and data heterogeneity design, which may not align with practical heterogeneous wireless networks. In this work, we advocate a new independent client sampling strategy to minimize the wall-clock training time of FL, while considering data heterogeneity and system heterogeneity in both communication and computation. We first derive a new convergence bound for non-convex loss functions with independent client sampling and then propose an adaptive bandwidth allocation scheme. Furthermore, we propose an efficient independent client sampling algorithm based on the upper bounds on the convergence rounds and the expected per-round training time, to minimize the wall-clock time of FL, while considering both the data and system heterogeneity. Experimental results under practical wireless network settings with real-world prototype demonstrate that the proposed independent sampling scheme substantially outperforms the current best sampling schemes under various training models and datasets.