Adaptive Federated Learning With Gradient Compression in Uplink NOMA

TL;DR

This paper proposes an adaptive federated learning scheme that combines gradient compression with NOMA to reduce latency in wireless uplinks, maintaining accuracy for mobile edge devices.

Contribution

It introduces a novel approach integrating gradient compression with NOMA in federated learning to improve uplink efficiency over wireless channels.

Findings

Significantly reduced aggregation latency

Achieved similar accuracy to traditional methods

Effective in power-limited wireless environments

Abstract

Federated learning (FL) is an emerging machine learning technique that aggregates model attributes from a large number of distributed devices. Several unique features such as energy saving and privacy preserving make FL a highly promising learning approach for power-limited and privacy sensitive devices. Although distributed computing can lower down the information amount that needs to be uploaded, model updates in FL can still experience performance bottleneck, especially for updates via wireless connections. In this work, we investigate the performance of FL update with mobile edge devices that are connected to the parameter server (PS) with practical wireless links, where uplink update from user to PS has very limited capacity. Different from the existing works, we apply non-orthogonal multiple access (NOMA) together with gradient compression in the wireless uplink. Simulation results show that our proposed scheme can significantly reduce aggregation latency while achieving similar accuracy.