A Discrete-Time Multi-Hop Consensus Protocol for Decentralized Federated Learning

This paper presents a Federated Learning (FL) algorithm that allows the decentralization of all FL solutions employ model-averaging procedure. The proposed proves to be capable attaining faster convergence rates and no performance loss against starting centralized implementation with reduced communication overhead compared existing consensus-based solutions. To this end, Multi-Hop consensus protocol, originally presented in scope dynamical system theory, leveraging on standard Lyapunov stability discussions, has been assure federation clients share same average model employing only information obtained from their m -step neighbours. Experimental results different topologies MNIST MedMNIST v2 datasets validate properties demonstrating drop, setting, about 1%.