Department of Computer, Control and Management Engineering

Federated Learning is a distributed learning solution for machine learning problems without the need of collecting the available data in a single centralized data centre. With the standard FL approaches, model training is performed locally and a centralized server collects and elaborates the trainable parameters of the local models: even if data are not shared, the presence of the centralized server still rises trust and security issues. In this work, we introduce the Decentralized Federated Learning (DECFEDAVG) algorithm, which aims at achieving complete decentralization by the lack of a coordination server, and compare its performance against the original federated learning algorithm Federated Averaging (FEDAVG) over the Nonintrusive Load Monitoring problem.