Samsung Research America’s Digital Health Team has thrown its hat into the health AI foundation model race, unveiling two systems built specifically for wearable biosignals. The models — xMAE and HiMAE — were presented as part of the company’s Connected Care vision, which Samsung discussed at its Health Forum during Galaxy Unpacked in July 2026. The pitch is preventive, personalized care powered by models that learn from heart activity, sleep, and movement data collected by smartwatches.
The standout is xMAE, short for Physiology-Aware Masked Cross-Modal Reconstruction for Biosignal Representation Learning. The model connects two cardiac signals that measure the heart through different mechanisms: ECG, which captures electrical activity but requires a user to pause for an active reading, and PPG, which tracks blood flow passively and continuously. Samsung compares the time lag between the two to hearing thunder after seeing lightning. By learning that temporal relationship, xMAE can reconstruct masked ECG segments from PPG data alone — potentially eliminating the need for manual ECG measurements during cardiovascular analysis. The pretraining used roughly 9,400 hours of combined ECG and PPG data.
The results are concrete. Samsung reports xMAE outperformed unimodal biosignal models and existing multimodal learning methods in 15 of 19 evaluation tasks, spanning cardiovascular disease prediction, abnormal test-result detection, and sleep-stage classification. The company also claims the learned features showed potential for use across sensor devices, body locations, and data-gathering environments. HiMAE, the second model, takes a different approach: it uses multiple encoders to analyze short and long data segments separately, letting the model pick the right time scale for a given task — fast-changing heartbeats versus slowly building sleep patterns.
Sharanya Desai, Head of Digital Health Algorithms at Samsung Research America, said the research “lays the technical groundwork for delivering health insights that are efficient, precise, and continuous through a health foundation model.” Subbu Venkatraman, Head of the Digital Health Research Lab, emphasized the significance of capturing both inter-signal relationships and temporal structure. Both papers have been accepted at major venues — xMAE at ICML and HiMAE at ICLR. Whether this translates into on-device features for Galaxy Watch users remains an open question, but Samsung is clearly betting that foundation models, not single-purpose algorithms, are the future of consumer health sensing.
