BiFedKD: New Framework for Federated ECG Monitoring
A new framework called BiFedKD (Bidirectional Federated Knowledge Distillation) has been proposed to improve ECG monitoring in IoMT networks. It addresses challenges of non-IID and long-tailed label distributions by using an aggregation-by-distillation pipeline with temperature scaling. The framework aims to reduce communication overhead while maintaining performance under data heterogeneity. Experiments were conducted on the MIT-BIH Arrhythmia dataset.
Key facts
- BiFedKD stands for Bidirectional Federated Knowledge Distillation
- It is designed for ECG monitoring in Internet of Medical Things (IoMT) networks
- Addresses non-IID and long-tailed label distributions
- Uses aggregation-by-distillation pipeline with temperature scaling
- Aims to reduce communication overhead compared to federated learning
- Tested on MIT-BIH Arrhythmia dataset
- Published on arXiv with ID 2605.14886
- Announcement type is new
Entities
Institutions
- arXiv