Distributed hierarchical deep optimization for federated learning in mobile edge computing

Zheng, Xiao, Shah, Syed Bilal Hussain, Bashir, Ali Kashif, Nawaz, Raheel and Rana, Omer ORCID: https://orcid.org/0000-0003-3597-2646 2022. Distributed hierarchical deep optimization for federated learning in mobile edge computing. Computer Communications 194 , pp. 321-328. 10.1016/j.comcom.2022.07.028

PDF - Accepted Post-Print Version Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (1MB)

Abstract

Deep learning has recently attracted great attention in many application fields, especially for big data analysis in the field of edge computing. Federated learning, as a promising machine learning technology, applies training data on distributed edge nodes to design shared learning systems to protect data privacy. Due to the system update in federated learning is at the expense of parameter exchange between edge nodes, it is extremely bandwidth consuming. A novel distributed hierarchical tensor depth optimization algorithm is proposed, which compresses the model parameters from the high-dimensional tensor space to a union of low-dimensional subspaces to reduce bandwidth consumption and storage demands of federated learning. In addition, an update method based on hierarchical tensor back propagation is developed by directly calculating the gradient of low-dimensional parameters to reduce the memory requirement and improve the training efficiency caused by edge node training. Finally, a large number of simulation experiments were performed to evaluate performance based on classical data sets with different local data distributions. Experimental results show that the proposed algorithm reduces the burden of communication bandwidth and the energy consumption of edge nodes.

Item Type:	Article
Date Type:	Publication
Status:	Published
Schools:	Computer Science & Informatics
Additional Information:	License information from Publisher: LICENSE 1: Title: This article is under embargo with an end date yet to be finalised.
Publisher:	Elsevier
ISSN:	0140-3664
Date of First Compliant Deposit:	29 July 2022
Date of Acceptance:	20 July 2022
Last Modified:	06 Nov 2023 14:11
URI:	https://orca.cardiff.ac.uk/id/eprint/151570

Citation Data

Cited 1 time in Scopus. View in Scopus. Powered By Scopus® Data

Actions (repository staff only)

Edit Item