From biological signals to health insights

LJUBLJANA, August 15, 2026, Artificial intelligence is increasingly being used to analyze biological signals measured by wearable devices such as smartwatches. By recognizing significant patterns in health data, including sleep, heart rate, and physical activity, artificial intelligence helps users better understand and manage their health. At a healthcare forum held during the Galaxy Unpacked event in July 2026, Samsung presented its vision for connected care, Connected Care, envisioning the next chapter of digital health, a future where care shifts from reactive treatment toward preventive, personal, and connected experiences supported by trustworthy healthcare innovations and partnerships throughout the entire healthcare ecosystem. One of the technologies driving these new consumer experiences is the healthcare foundation model. Researchers from Samsung's Digital Health team at Samsung Research America (SRA) are developing new AI technologies that continuously understand human health status based on biological signals, create health insights, and provide appropriate health guidance. Samsung researchers recently introduced two foundational models designed using wearable device data: xMAE (physiological-aware masked multimodal reconstruction for learning representation of biological signals), which learns the temporal relationship between different biological signals, and HiMAE (hierarchical masked self-coder), which understands health patterns at different time scales within time series data from wearable devices. Both studies demonstrate progress in healthcare models of artificial intelligence capable of better understanding physiological conditions and the temporal structure of biological signals. Samsung’s work on the xMAE and HiMAE models has been accepted at the International Conference on Machine Learning and the International Conference on Learning Representations, highlighting the significance of these research efforts. Healthcare foundation models, reading body signals: why they are important for health A healthcare foundation model is an AI model that uses self-supervised learning to learn important features from unlabeled data of biological signals. After prior training on extensive healthcare data, these models can be used to perform a wide range of subsequent healthcare tasks, including analyzing biological signals, developing new or improved biological indicators, and predicting health issues. The AI model xMAE: continuous insight into heart function through wearable devices An electrocardiogram on wearable devices directly measures the electrical activity of the heart and is useful for measuring heart rate and heart rate variability, identifying irregular heart rhythms, and detecting risks for conditions such as atrial fibrillation. An electrocardiogram is highly accurate but usually requires users to take a break and actively measure it. On the other hand, photoplethysmography indirectly measures heart function by detecting changes in blood flow and can be passively and continuously measured with sensors in wearable devices such as smartwatches. Both signals originate from the same cardiac activity, but they appear with a certain time delay, similar to how thunder is heard after seeing lightning. xMAE is a pre-training framework for biological signals designed to learn the temporal relationship between both signals by reconstructing masked parts of the electrocardiogram signal using the photoplethysmography signal, which is easier and continuously measurable. As a result, features related to cardiovascular health can be more accurately analyzed using photoplethysmography without the need for separate manual measurements of the electrocardiogram. Researchers trained the xMAE model using approximately 9,400 hours of electrocardiogram and photoplethysmography data. The model surpassed unimodal models of biological signals and existing ones.