Scheduling for Ground-Assisted Federated Learning in LEO Satellite Constellations

TL;DR

This paper proposes a scheduler for ground-assisted federated learning in LEO satellite constellations that leverages predictable satellite-ground station contact times to significantly enhance training convergence speed.

Contribution

It introduces a novel scheduler tailored for satellite FL that exploits contact time predictability to reduce model staleness and accelerate convergence.

Findings

Convergence speed improved by a factor of three.

Scheduler effectively exploits satellite-ground contact predictability.

Enhances federated learning efficiency in LEO satellite networks.

Abstract

Distributed training of machine learning models directly on satellites in low Earth orbit (LEO) is considered. Based on a federated learning (FL) algorithm specifically targeted at the unique challenges of the satellite scenario, we design a scheduler that exploits the predictability of visiting times between ground stations (GS) and satellites to reduce model staleness. Numerical experiments show that this can improve the convergence speed by a factor three.