Volume 19 Issue 4
August  2022
Turn off MathJax
Article Contents
Alessandro Giuseppi, Sabato Manfredi, Antonio Pietrabissa. A Weighted Average Consensus Approach for Decentralized Federated Learning. Machine Intelligence Research, vol. 19, no. 4, pp.319-330, 2022. https://doi.org/10.1007/s11633-022-1338-z
 Citation: Alessandro Giuseppi, Sabato Manfredi, Antonio Pietrabissa. A Weighted Average Consensus Approach for Decentralized Federated Learning. Machine Intelligence Research, vol. 19, no. 4, pp.319-330, 2022.

# A Weighted Average Consensus Approach for Decentralized Federated Learning

##### doi: 10.1007/s11633-022-1338-z
• Author Bio:

Alessandro Giuseppi received the M. Sc. degree in control engineering and the Ph. D. degree in automatica from University of Rome La Sapienza, Italy in 2016 and 2019, respectively. Since 2016, he has participated in 6 other EU and National research projects, mainly in the fields of control systems and artificial intelligence. He is an assistant professor in automatica at Department of Computer, Control, and Management Engineering Antonio Ruberti (DIAG), University of Rome La Sapienza, Italy. Currently, he is the scientific coordinator of the ESA-funded research project ARIES, related to wildfire emergency management and work package leader in the EU-Korea H2020 project 5G-ALLSTAR. Since 2020, he is serving as associate editor for the International Journal of Control, Automation and Systems (Springer). His main research activities are in the fields of network control and intelligent systems, where he published about 50 papers in international journals and conferences. His research interests include intelligent systems and automatic control, with emphasis on smart networks. E-mail: giuseppi@diag.uniroma1.it (Corresponding author) ORCID iD: 0000-0001-5503-8506

Sabato Manfredi received the M. Sc. degree in electronics engineering and the Ph. D. degree in computer science and automatica from University of Naples Federico II, Italy in 2001 and 2004, respectively. He is currently an associate professor of automatic control with Department of Electrical Engineering and Information Technology, University of Naples Federico II, Italy. He has been a visiting academic with the Control and Power Group, Electrical and Electronic Engineering Department, Imperial College London, UK since 2012. He has been a visiting professor with School of Mathematical Sciences, Queen Mary, UK, during 2017–2018. He has authored/coauthored more than 90 scientific publications including 18 single-author papers and the monograph: Multilayer Control of Networked Cyber-Physical Systems. Application to Monitoring, Autonomous and Robot Systems (Advances in Industrial Control Series, Springer, 2017). He collaborates with many international universities and companies, holds European patent, is a proponent member of an academic spin-off, and is involved in a range of academic, industrial, and consulting projects. His research interests include automatic control with a special emphasis on nonlinear and complex networks, distributed control and optimization, sensor/drone networks, and new technologies/algorithms for smart city and cyber–physical systems. E-mail: sabato.manfredi@unina.it

Antonio Pietrabissa received the M. Sc. degree in electronics engineering and the Ph. D. degree in systems engineering from University of Rome “La Sapienza”, Italy in 2000 and 2004, respectively, and where he teaches automatic control and process automation. Since 2000, he has participated in about 25 EU and National research projects. He is associate professor at Department of Computer, Control, and Management Engineering “Antonio Ruberti” (DIAG), University of Rome “La Sapienza”, Italy. Currently, he is the coordinator of the project ARIES on fire emergency prevention, funded by ESA, and the scientific responsible of the research projects 5G-ALLSTARS on 5G communications, funded within the H2020 Europe-South Korea cooperation, and FedMedAI on medical applications of federated learning. He serves as associate editor for Control Engineering Practice (Elsevier) and for IEEE Transactions on Automation Science and Engineering. He is author of more than 50 journal papers and 80 conference papers. His research interest is the application of systems and control theory to the analysis and control of networks. E-mail: pietrabissa@diag.uniroma1.it

• Accepted Date: 2022-05-05
• Publish Date: 2022-08-01
• Federated learning (FedL) is a machine learning (ML) technique utilized to train deep neural networks (DeepNNs) in a distributed way without the need to share data among the federated training clients. FedL was proposed for edge computing and Internet of things (IoT) tasks in which a centralized server was responsible for coordinating and governing the training process. To remove the design limitation implied by the centralized entity, this work proposes two different solutions to decentralize existing FedL algorithms, enabling the application of FedL on networks with arbitrary communication topologies, and thus extending the domain of application of FedL to more complex scenarios and new tasks. Of the two proposed algorithms, one, called FedLCon, is developed based on results from discrete-time weighted average consensus theory and is able to reconstruct the performances of the standard centralized FedL solutions, as also shown by the reported validation tests.

• 1 Actually, in practice, the local weight update is performed iteratively over $E$ training epochs using a variation of gradient descent (mini-batch gradient descent) that splits $D_i$ into a set of mini-batches. Equation (3) exemplifies the update rule with E = 1 and over the complete dataset, whereas the pseudo-codes report the mini-batch multi-epoch version of the algorithms.

### Catalog

###### 通讯作者: 陈斌, bchen63@163.com
• 1.

沈阳化工大学材料科学与工程学院 沈阳 110142

Figures(8)  / Tables(1)

用微信扫码二维码

分享至好友和朋友圈