Optimisation of federated learning settings under statistical heterogeneity variations

TL;DR

This paper empirically analyzes how federated learning performance varies with statistical heterogeneity, proposing strategies and guidelines to optimize model training across diverse data distributions.

Contribution

It introduces a systematic data partitioning method, a heterogeneity metric, and provides empirical guidelines for selecting FL parameters and aggregators based on data characteristics.

Findings

Optimal FL parameters vary with data heterogeneity levels.

Certain aggregators perform better under specific heterogeneity conditions.

Guidelines improve FL model performance across diverse datasets.

Abstract

Federated Learning (FL) enables local devices to collaboratively learn a shared predictive model by only periodically sharing model parameters with a central aggregator. However, FL can be disadvantaged by statistical heterogeneity produced by the diversity in each local devices data distribution, which creates different levels of Independent and Identically Distributed (IID) data. Furthermore, this can be more complex when optimising different combinations of FL parameters and choosing optimal aggregation. In this paper, we present an empirical analysis of different FL training parameters and aggregators over various levels of statistical heterogeneity on three datasets. We propose a systematic data partition strategy to simulate different levels of statistical heterogeneity and a metric to measure the level of IID. Additionally, we empirically identify the best FL model and key parameters for datasets of different characteristics. On the basis of these, we present recommended guidelines for FL parameters and aggregators to optimise model performance under different levels of IID and with different datasets