Federated Learning in Satellite Constellations

Federated learning (FL) has recently emerged as a distributed machine paradigm for systems with limited and intermittent connectivity. This paper presents the new context brought to FL by satellite constellations, where connectivity patterns are significantly different from ones observed in conventional terrestrial FL. The focus is on large constellations low Earth orbit (LEO), each satellites participates data-driven task using locally stored dataset. scenario motivated trend towards mega of interconnected small LEO integration artificial intelligence satellites. We propose classification based communication capabilities satellites, constellation design, location parameter server. A comprehensive overview current state-of-the-art this field provided unique challenges opportunities discussed. Finally, we outline several open research directions present some future perspectives topic.