Pervasive new era applications are expected to involve massive amount of data to implement intelligent distributed frameworks based on machine learning, supported by sixth generation (6G) networks technology to offer fast and reliable communications. Federated Learning (FL) is rapidly emerging as promising privacy-preserving solution to train machine learning models in a distributed fashion. However, users are often not too inclined to take part in the learning process without receiving compensation. Hence, to overcome this drawback, the functional integration of a proper devices incentive mechanism with an efficient approach for the devices selection in a same FL framework becomes essential. In this regard, this paper proposes a FL framework involving a one-side matching theory-based incentive mechanism to select and encourage users to take part of the process with the aim at minimizing the FL process convergence time and maximizing the users profit. Furthermore, this paper faces with the possibility to overcome bad communication link conditions by resorting to device-to-device communications among users in order to lower the energy wasted and improve the convergence time of the FL process. In particular, an echo- state-network, running in local at each user site, has been considered to forecast channel conditions in a reliable manner. Performance evaluation has highlighted the improvements in convergence time and energy consumption of the proposed FL framework in comparison with conventional approaches, hence, highlighting its suitability for applications in the upcoming 6G networks

A D2D-Aided Federated Learning Scheme with Incentive Mechanism in 6G Networks / ROMANO FANTACCI; Benedetta Picano. - In: IEEE ACCESS. - ISSN 2169-3536. - STAMPA. - 11:(2023), pp. 107-117. [10.1109/ACCESS.2022.3232440]

A D2D-Aided Federated Learning Scheme with Incentive Mechanism in 6G Networks

ROMANO FANTACCI;Benedetta Picano
2023

Abstract

Pervasive new era applications are expected to involve massive amount of data to implement intelligent distributed frameworks based on machine learning, supported by sixth generation (6G) networks technology to offer fast and reliable communications. Federated Learning (FL) is rapidly emerging as promising privacy-preserving solution to train machine learning models in a distributed fashion. However, users are often not too inclined to take part in the learning process without receiving compensation. Hence, to overcome this drawback, the functional integration of a proper devices incentive mechanism with an efficient approach for the devices selection in a same FL framework becomes essential. In this regard, this paper proposes a FL framework involving a one-side matching theory-based incentive mechanism to select and encourage users to take part of the process with the aim at minimizing the FL process convergence time and maximizing the users profit. Furthermore, this paper faces with the possibility to overcome bad communication link conditions by resorting to device-to-device communications among users in order to lower the energy wasted and improve the convergence time of the FL process. In particular, an echo- state-network, running in local at each user site, has been considered to forecast channel conditions in a reliable manner. Performance evaluation has highlighted the improvements in convergence time and energy consumption of the proposed FL framework in comparison with conventional approaches, hence, highlighting its suitability for applications in the upcoming 6G networks
2023
11
107
117
ROMANO FANTACCI; Benedetta Picano
File in questo prodotto:
File Dimensione Formato  
A_D2D-Aided_Federated_Learning_Scheme_With_Incentive_Mechanism_in_6G_Networks.pdf

accesso aperto

Tipologia: Pdf editoriale (Version of record)
Licenza: Open Access
Dimensione 907.58 kB
Formato Adobe PDF
907.58 kB Adobe PDF

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

Utilizza questo identificatore per citare o creare un link a questa risorsa: https://hdl.handle.net/2158/1294599
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 2
  • ???jsp.display-item.citation.isi??? 1
social impact