Robust Federated Learning in Wireless Channels with Transmission Outage and Quantization Errors

Abstract

Federated learning (FL) has been recognized as a viable distributed learning paradigm which trains a machine learning model collaboratively with massive mobile devices in the wireless edge while protecting user privacy. Although various communication schemes have been proposed to reduce the training latency in order to expedite the FL learning process, most of them have assumed that the wireless channels are ideal which provide reliable and lossless communication links between the server and mobile clients. Unfortunately, the performance of FL highly depends on the quality of wireless channels. In this paper, we consider FL in non-ideal wireless channels with transmission outage (TO) and quantization errors (QE). We analytically show that both TO and QE have significant impacts on the FL and can lead the FL algorithm to a biased solution if the clients have heterogeneous TO probabilities. In view of this, we propose a robust FL scheme which performs joint bandwidth and quantization bit allocation across clients to achieve robust FL performance in such non-ideal wireless channels. Extensive experimental results are presented to show the superior performance of the proposed scheme for a deep learning-based classification task with training latency constraints.