Samsung health AI models analyse wearable biosignal data

Details

Samsung's Digital Health Team unveiled two foundation models for wearable biosignal data at the Health Forum during Galaxy Unpacked in July 2026. Both use self-supervised learning on unlabeled biosignal data to build transferable representations for downstream health tasks — addressing the practical scarcity of labeled health data. xMAE (Physiology-Aware Masked Cross-Modal Reconstruction) learns temporal relationships between different biosignal types, specifically linking continuous PPG data (optical heart rate from the wrist) to ECG electrical activity. The goal is enabling ECG-quality cardiac insights from cheaper, always-on PPG sensors in smartwatches. Accepted to ICML. HiMAE (Hierarchical Masked Autoencoder) learns health patterns across multiple time scales — short windows for beat-to-beat variation and long windows for circadian-scale rhythms — from a single pretrained model. It supports classification, numerical prediction, and data generation tasks. Accepted to ICLR. Samsung frames both models as foundational infrastructure for its 'Connected Care' vision: preventive, personalized health monitoring running on Galaxy watches with limited sensors and on-device compute. The dual ICML/ICLR acceptances provide peer-reviewed validation that the underlying research is technically credible.