Unsupervised Data Splitting Scheme for Federated Edge Learning in IoT Networks

TL;DR

This paper introduces an unsupervised data-aware splitting and node selection scheme for federated edge learning in IoT networks, reducing energy consumption and communication rounds while maintaining training quality.

Contribution

It presents a novel unsupervised data partitioning and heuristic node selection method that improves energy efficiency and reduces communication in FEEL systems.

Findings

Significantly reduces energy consumption compared to vanilla FEEL.

Decreases the number of communication rounds needed for training.

Enhances training performance by selecting high-quality data and nodes.

Abstract

Federated Edge Learning (FEEL) is a promising distributed learning technique that aims to train a shared global model while reducing communication costs and promoting users' privacy. However, the training process might significantly occupy a long time due to the nature of the used data for training, which leads to higher energy consumption and therefore impacts the model convergence. To tackle this issue, we propose a data-driven federated edge learning scheme that tends to select suitable participating nodes based on quality data and energy. First, we design an unsupervised data-aware splitting scheme that partitions the node's local data into diverse samples used for training. We incorporate a similarity index to select quality data that enhances the training performance. Then, we propose a heuristic participating nodes selection scheme to minimize the communication and computation energy consumption, as well as the amount of communication rounds. The obtained results show that the proposed scheme substantially outperforms the vanilla FEEL in terms of energy consumption and the number of communication rounds.