Towards Family-Grouped Hierarchical Federated Learning on Sub-5KB Models: A Feasibility Study of Privacy-Preserving ECG Monitoring for Ultra-Resource-Constrained Wearables
A new study proposes a Family-Grouped Hierarchical Federated Learning (Family-FL) model for privacy-preserving ECG monitoring on ultra-resource-constrained wearables. This model significantly reduces communication overhead while maintaining accuracy in detecting arrhythmias. The research demonstrates the feasibility of implementing federated learning on low-power microcontrollers, although it acknowledges limitations such as lack of hardware deployment and single-dataset validation.
- ▪Cardiovascular disease is the leading cause of death globally, and continuous ECG monitoring can help in early detection of arrhythmias.
- ▪The Family-FL model reduces communication volume by 76.7% compared to traditional methods while achieving 91.9% accuracy.
- ▪The Tiny CNN-LSTM architecture developed for this study is designed to fit within the constraints of low-power microcontrollers.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Machine Learning arXiv:2605.18862 (cs) [Submitted on 15 May 2026] Title:Towards Family-Grouped Hierarchical Federated Learning on Sub-5KB Models: A Feasibility Study of Privacy-Preserving ECG Monitoring for Ultra-Resource-Constrained Wearables Authors:Hangyu Wu View a PDF of the paper titled Towards Family-Grouped Hierarchical Federated Learning on Sub-5KB Models: A Feasibility Study of Privacy-Preserving ECG Monitoring for Ultra-Resource-Constrained Wearables, by Hangyu Wu View PDF HTML (experimental) Abstract:Cardiovascular disease remains the leading cause of death worldwide, and early detection of arrhythmias through continuous ECG monitoring on wearable devices can prevent life-threatening events.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
