Fast-tracking the deep residual network training for arrhythmia classification by leveraging the power of dynamical systems

Bohoran, T.A., Kampaktsis, P.N., McCann, G.P. and Giannakidis, A. ORCID: 0000-0001-7403-923X, 2024. Fast-tracking the deep residual network training for arrhythmia classification by leveraging the power of dynamical systems. In: 17th International Conference on Signal-Image Technology and Internet-Based Systems (SITIS 2023) proceedings. IEEE. ISBN 9798350370928

Preview

Text 1878526_Giannakidis.pdf - Post-print Download (3MB) | Preview

Abstract

Arrhythmia, characterised by irregular heartbeats, poses significant health risks. Residual networks, a subset of deep learning architectures, have emerged as a potent tool in detecting electrocardiogram (ECG) signal anomalies. However, the enhanced accuracy and capabilities afforded by increasing network depth in these models come at the cost of heightened computational demands, posing a considerable challenge to their practical applicability. Addressing this critical bottleneck, this paper presents a methodology for the resource-economical development of machine learning-enabled systems for arrhythmia detection. The proposed methodology, grounded in the dynamical systems perspective of residual networks, initiates the training process with a shallow network, and then progressively augments its depth. We validate the method rigorously on the PhysioNet’s MIT-BIH arrhythmia dataset using heartbeat spectrograms as training inputs. The results show that the proposed training necessitates a minimum of 39.47% fewer parameters per epoch, when compared with the conventional vanilla training, a feat achieved without sacrificing, and in fact potentially enhancing, the overall performance. Our findings suggest the methodology not only drastically reduces training time but also promises significant savings in energy consumption and environmental costs, offering a glimpse into a future of more sustainable and resource-efficient machine learning developments in arrhythmia detection.

Item Type:	Chapter in book
Description:	Paper presented at 17th IEEE International Conference on Signal-Image Technology and Internet-Based Systems (SITIS), Bangkok, Thailand, 8-10 November 2023.
Creators:	Bohoran, T.A., Kampaktsis, P.N., McCann, G.P. and Giannakidis, A.
Publisher:	IEEE
Date:	21 March 2024
ISBN:	9798350370928
Identifiers:	Number Type 10.1109/sitis61268.2023.00035 DOI 1878526 Other
Divisions:	Schools > School of Science and Technology
Record created by:	Laura Ward
Date Added:	26 Mar 2024 09:39
Last Modified:	26 Mar 2024 09:44
URI:	https://irep.ntu.ac.uk/id/eprint/51156

Actions (login required)

Edit View

Views

Downloads