Conveners
Federated Learning and efficient learning for Edge Based Biomedical Signal Analysis
- Paolo Cacace (Sapienza Universita e INFN, Roma I (IT))
Description
My contribution introduces a federated and efficient deep learning framework for biomedical signal analysis, with a focus on ECG signal processing. The approach leverages technologies originally developed at CERN for time-series anomaly detection in industrial systems, adapting them to the healthcare domain to ensure scalability, efficiency, and data privacy. A hybrid architecture combining convolutional neural networks and Transformer attention layers was used to capture both fine-grained waveform patterns and long-term temporal dependencies in multichannel physiological signals.
The framework has been successfully tested on edge devices, demonstrating the feasibility of real-time inference under resource constraints. To enable personalized healthcare while preserving patient data privacy, further development will see those model federated using CAFEIN®, a federated learning platform developed at CERN. The objective is to demonstrate how cross domain knowledge transfer and efficient AI can drive the development of next-generation biomedical solutions operating at the edge.