🤯 AI Heart Secrets: Future of Health Unlocked 💖

THE RISE OF HEALTH FOUNDATION MODELS
Samsung Research America’s Digital Health Team is pioneering the development of two AI foundation models designed to analyze data gleaned from wearable biosignals. This research centers on data collected by smartwatches, specifically focusing on heart activity, sleep patterns, and physical activity levels. The company’s “Connected Care” vision, unveiled at the Health Forum during Galaxy Unpacked in July 2026, envisions a future of preventative, personalized, and connected healthcare, fueled by health technology and strategic healthcare partnerships. These foundation models represent a key component of Samsung’s ambitions in this evolving landscape.

SHARANYA DESAI’S VISION
Head of Digital Health Algorithms at Samsung Research America, Sharanya Desai, emphasized the significance of this work: “This research lays the technical groundwork for delivering health insights that are efficient, precise, and continuous through a health foundation model.” Desai highlighted Samsung’s ongoing commitment to developing and advancing these models, adaptable to a wide range of biosignals and health features, capable of operating on-device with limited sensors and computing resources.

XMAE: PHYSIOLOGY-AWARE RECONSTRUCTION
The first model, xMAE (Physiology-Aware Masked Cross-Modal Reconstruction for Biosignal Representation Learning), employs self-supervised learning to identify key features within unlabeled biosignal data. Samsung’s approach involves pretraining these models on extensive health datasets, enabling a single model to support diverse downstream tasks, including biosignal analysis, biomarker development, and health issue prediction. The xMAE model specifically connects two cardiac signals – PPG and ECG – measuring related activity through distinct mechanisms.

HIMATE: LEARNING HEALTH PATTERNS ACROSS TIME
The second model, HiMAE (Hierarchical Masked Autoencoder), learns health patterns across multiple time scales in wearable time-series data. This approach allows the model to analyze data at both short and long intervals, supporting classification, numerical prediction, and data generation. HiMAE utilizes multiple encoders to analyze short and long data segments separately, enabling it to identify the optimal time scale for a specific health task, such as heart-rate analysis or sleep prediction.

THE SCIENCE OF SIGNAL INTERPRETATION
Both xMAE and HiMAE address different aspects of wearable data analysis. xMAE’s design focuses on analyzing cardiovascular health through continuously measured PPG data, eliminating the need for separate manual ECG measurements. The model’s pretraining utilized approximately 9,400 hours of ECG and PPG data. Subbu Venkatraman, Head of the Digital Health Research Lab at Samsung Research America, stated, “Biosignals are inherently dynamic, with unique time-varying physiological properties. The key contribution of this research lies in proving the viability of health foundation models capable of capturing both the inter-signal relationships and their underlying temporal structures.”

PERFORMANCE AND POTENTIAL
Samsung reports that xMAE outperformed unimodal biosignal models and existing multimodal learning methods in 15 of 19 evaluation tasks. These tasks included cardiovascular disease prediction, abnormal test-result detection, and sleep-stage classification. The learned features demonstrated potential for use across various sensor devices, body locations, and data-gathering environments.

TIME-SCALING WITH HIMATE
HiMAE analyzes wearable data across multiple time scales. Wearable data can contain different information over different periods. Short segments reveal fast-changing signals like heartbeats, while longer segments reveal patterns that build over time, such as sleep or physical activity. The model reconstructs masked portions of wearable data, learning patterns even with limited labeled data. This enables classification, numerical prediction, and data generation from a single pretrained system.

SPEED AND EFFICIENCY
Notably, HiMAE achieved high performance with a smaller model than existing models, and can produce results in less than one millisecond on a smartwatch-class central processing unit. This highlights the potential for real-time health insights directly from wearable devices.

FUTURE DIRECTIONS
Samsung’s research underscores the potential of health foundation models to revolutionize healthcare by leveraging the continuous stream of data generated by wearable devices. The company remains committed to advancing foundational health AI research and translating it into healthcare solutions that meaningfully improve people’s health and wellbeing.