Federated learning is a popular framework that enables harvesting edge resources' computational power to train a machine learning model distributively. However, it is not always feasible or profitable to have a centralized server that controls and synchronizes the training process. In this paper, we consider the problem of training a machine learning model over a network of nodes in a fully decentralized fashion. In particular, we look for empirical evidence on how sensitive is the training process for various network characteristics and communication parameters. We present the outcome of several simulations conducted with different network topologies, datasets, and machine learning models.

Impact of Network Topology on the Convergence of Decentralized Federated Learning Systems

Dazzi, Patrizio;Ferrucci, Luca;
2021-01-01

Abstract

Federated learning is a popular framework that enables harvesting edge resources' computational power to train a machine learning model distributively. However, it is not always feasible or profitable to have a centralized server that controls and synchronizes the training process. In this paper, we consider the problem of training a machine learning model over a network of nodes in a fully decentralized fashion. In particular, we look for empirical evidence on how sensitive is the training process for various network characteristics and communication parameters. We present the outcome of several simulations conducted with different network topologies, datasets, and machine learning models.
2021
978-1-6654-2744-9
File in questo prodotto:
File Dimensione Formato  
Impact_of_Network_Topology_on_the_Convergence_of_Decentralized_Federated_Learning_Systems.pdf

non disponibili

Tipologia: Versione finale editoriale
Licenza: NON PUBBLICO - accesso privato/ristretto
Dimensione 1.18 MB
Formato Adobe PDF
1.18 MB Adobe PDF   Visualizza/Apri   Richiedi una copia

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11568/1143882
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 2
  • ???jsp.display-item.citation.isi??? 1
social impact