Domain Adaptation for Online ECG Monitoring

摘要: Successful ECG monitoring algorithms often rely on learned models to describe the heartbeats morphology. Unfortunately, when heart rate increases get transformed, and a model that can properly of specific user in resting conditions might not be appropriate for same during everyday activities. We by dictionaries yielding sparse representations propose novel domain-adaptation solution which transforms user-specific according rate. In particular, we learn suitable linear transformations from large dataset containing tracings, show these successfully adapt changes. Remarkably, used multiple users different sensing apparatus. investigate implications our findings wearable devices, present an efficient implementation anomaly-detection algorithm leveraging such transformations.