From Biosignals to Health Insights: Samsung Research’s Work on Health Foundation Models

Samsung Research has developed xMAE and HiMAE, two health foundation models designed to interpret biosignals from wearables like smartwatches. xMAE reconstructs masked ECG data using PPG signals to enable continuous cardiac monitoring without requiring separate manual ECG measurements. The model was pretrained on approximately 9,400 hours of ECG and PPG data and outperformed existing methods in 15 of 19 evaluation tasks, including cardiovascular disease prediction and sleep-stage classification.

HiMAE processes wearable data across multiple time scales, analyzing short segments for tasks like heart rate monitoring and longer segments for sleep prediction. The model uses self-supervised learning to reconstruct masked data, improving efficiency and reducing reliance on labeled datasets. HiMAE achieved high performance while being smaller than existing models and can produce results in less than one millisecond on a smartwatch-class CPU.

Both models were accepted to the International Conference on Machine Learning (ICML) and the International Conference on Learning Representations (ICLR), respectively, underscoring their significance in advancing health AI research. The work aligns with Samsung’s Connected Care vision, which aims to shift digital health toward preventive, personalized, and connected experiences through trusted innovation and ecosystem partnerships.

The xMAE model leverages the temporal relationship between ECG and PPG signals—similar to the delay between lightning and thunder—to generate health insights without manual intervention. Together, these models demonstrate the potential for on-device health foundation models to deliver precise, continuous, and personalized health insights while generalizing across multiple health tasks from a single pretrained model.