The integration of artificial intelligence in analyzing biosignals from wearable technology, such as smartwatches, is transforming personal health management. AI’s capability to identify patterns in biometric data including sleep cycles, heart rates, and physical activity is empowering users to gain deeper insights into their health.
Samsung’s Vision for Connected Health
During the Galaxy Unpacked event in July 2026, Samsung unveiled its Connected Care vision, which emphasizes a shift from reactive healthcare to preventative, personalized, and interconnected health solutions. This vision is being realized through health foundation models that leverage innovative AI technologies and strategic partnerships within the healthcare industry.
Development of Health Foundation Models
The Digital Health Team at Samsung Research America (SRA) is at the forefront of developing AI technologies that interpret biosignals, generate health insights, and provide relevant health guidance. Recently, they introduced two foundation models that utilize wearable data: xMAE and HiMAE.
- xMAE (Physiology-Aware Masked Cross-Modal Reconstruction): This model focuses on understanding the temporal relationships between different biosignals. It reconstructs masked ECG signals by learning from more continuously available PPG signals.
- HiMAE (Hierarchical Masked Autoencoder): Specializes in recognizing health patterns across multiple time scales, allowing the extraction of key insights from time-series data collected by wearable devices.
Both models have received recognition by being accepted at prestigious conferences, including the International Conference on Machine Learning (ICML) and the International Conference on Learning Representations (ICLR).
Understanding Health Foundation Models
Health foundation models utilize self-supervised learning to discern significant features from unlabeled biosignal data. After pretraining on large datasets, these models can perform a variety of health-related tasks, such as biosignal analysis and health issue prediction, thereby contributing to the development of new biomarkers.
xMAE AI Model: Enhancing Cardiovascular Monitoring
Traditional ECG measurements on wearables, though accurate, require user intervention. In contrast, PPG signals can be passively and continuously monitored. The xMAE model leverages this by reconstructing ECG signals using PPG data, enabling more precise cardiovascular analyses without manual measurements.
Pretrained on approximately 9,400 hours of data, xMAE outperformed existing models across 15 of 19 evaluation tasks, showcasing its versatility across various devices and environments.
HiMAE AI Model: Multiscale Biosignal Analysis
HiMAE excels in analyzing biosignals across different time scales. By utilizing multiple encoders, it discerns information pertinent to health tasks, such as heart rate analysis over brief periods, or long-term patterns like sleep habits.
The model is lightweight yet highly efficient, offering real-time analysis directly on devices such as smartwatches, eliminating the need for cloud-based processing.
The Future of Digital Health
xMAE and HiMAE signify a significant leap in AI’s ability to interpret complex biosignals. Their development represents a move towards delivering comprehensive, continuous, and personalized health insights, reinforcing Samsung’s commitment to advancing connected care through innovative AI solutions.