Samsung Research is releasing two new AI models trained on data from wearable sensors. Named xMAE and HiMAE, the foundation models are built to pull health insight from signals like heart rate, sleep, and activity.
The first model, xMAE, focuses on how different biosignals relate to one another. In cardiac use it learns how ECG and PPG readings from the same heartbeat line up over time, which could let wearables track the heart continuously instead of on demand. The second model, HiMAE, searches for health patterns that stretch across multiple time scales in the data.
Both rely on self-supervised learning, pulling structure from large piles of unlabeled biosignal data. After pretraining they can handle tasks like biosignal analysis, biomarker development, and health issue prediction. The work comes from the Digital Health team at Samsung Research America and was accepted to ICML and ICLR, two top machine learning conferences.
The launch fits Samsung’s broader Connected Care push, previewed at Galaxy Unpacked in July 2026, which frames care as preventive and personalized. For the wearables industry, the research points beyond step counting toward models that interpret what the body’s signals mean.
