federated learning fl

poisoning attacks

global model

earth observation

communication overhead

local differential privacy

decentralized federated

leo satellite

IEEE Wireless Communications Letters

In Low Earth Orbit (LEO) mega constellations, there are relevant use cases,\nsuch as inference based on satellite imaging, in which a large number of\nsatellites collaboratively train a machine learning model without sharing their\nlocal datasets. To address this problem, we propose a new set of algorithms\nbased on Federated learning (FL), including a novel asynchronous FL procedure\nbased on FedAvg that exhibits better robustness against heterogeneous scenarios\nthan the state-of-the-art. Extensive numerical evaluations based on MNIST and\nCIFAR-10 datasets highlight the fast convergence speed and excellent asymptotic\ntest accuracy of the proposed method.\n

Ground-Assisted Federated Learning in LEO Satellite Constellations