Samsung’s New AI Models Learn From Your Smartwatch to Predict Health Issues
Samsung just dropped something interesting. They’ve built two AI models that learn from wearable data like heart activity, sleep patterns, and physical activity. And the goal is pretty ambitious.
Let me break down what they’re doing.
What’s Going On?
Samsung Research America’s Digital Health Team presented two AI foundation models designed to learn from biosignals captured by smartwatches. We’re talking about data from Galaxy Watches and other wearables.
The company talked about their “Connected Care” vision at the Health Forum during Galaxy Unpacked in July 2026. They’re describing a future of preventive, personalized, and connected care supported by health technology and partnerships.
Basically, they want your smartwatch to predict health issues before they become problems.
The Two Models: xMAE and HiMAE
Samsung built two different models for two different jobs.
xMAE (Physiology-Aware Masked Cross-Modal Reconstruction for Biosignal Representation Learning)
This model learns the relationship between different biosignals. Specifically, it connects PPG data to ECG signals.
Here’s the deal. ECGs measure the heart’s electrical activity directly. They’re useful for measuring heart rate, heart rate variability, and identifying abnormal rhythms like atrial fibrillation. But they require you to actively take a measurement.
PPG, on the other hand, detects changes in blood flow and runs passively through sensors in wearables. It’s always on.
xMAE learns the relationship between these two signals by reconstructing masked parts of an ECG signal from PPG data. Think of it like hearing thunder after seeing lightning. Both come from the same event, just at different times.
The goal? Analyze cardiovascular health features through continuously measured PPG data without needing separate manual ECG measurements.
HiMAE (Hierarchical Masked Autoencoder)
This model learns health patterns across multiple time scales.
Wearable data carries different information over different time periods. Short segments show fast-changing signals like heartbeats. Longer segments reveal patterns that build over time, like sleep or physical activity.
HiMAE uses multiple encoders to analyze short and long data segments separately. This lets it identify what time scale is needed for a specific health task. Heart rate analysis and sleep prediction can draw on different parts of the time-series data.
The interesting part? HiMAE can produce results in less than one millisecond on a smartwatch-class CPU. That means the analysis happens on your device, not on cloud servers. No internet connection needed.
The Data
The models were trained on serious amounts of data.
xMAE pretraining used about 9,400 hours of ECG and PPG data. That’s nearly 400 days of continuous heart data.
And the results? xMAE outperformed unimodal biosignal models and existing multimodal learning methods in 15 of 19 evaluation tasks. Those tasks covered:
- Cardiovascular disease prediction
- Abnormal test result detection
- Sleep stage classification
The company also says the learned features showed potential for use across different sensor devices, body locations, and data-gathering environments.
What This Actually Means
Here’s the bottom line.
Samsung is building AI that can run on your smartwatch and analyze your health data in real time, without sending anything to the cloud. That means faster analysis and better privacy.
The models learn from unlabeled data. That’s important because labeled health data is expensive and time-consuming to produce. By using self-supervised learning, Samsung can train models on massive amounts of raw biosignal data.
The goal is to create a system that can:
- Extract diagnostic markers from physiological streams
- Run predictive health classifications
- Generate user guidance from consumer hardware
All without continuous server connectivity.
What the Executives Said
Sharanya Desai, Head of Digital Health Algorithms at Samsung Research America, put it this way:
“This research is significant because it 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, added:
“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.”
He also said they’re committed to “translating it into healthcare solutions that meaningfully improve people’s health and wellbeing.”
The Bigger Picture
This is part of a larger trend.
Tech companies are pushing deeper into health. Samsung, Apple, Google, and others are all building health features into their wearables. The difference here is the AI approach.
By building foundation models that learn from unlabeled biosignal data, Samsung is essentially creating a base layer of health intelligence that can be applied to many different tasks. One model that can handle heart rate analysis, sleep prediction, and disease risk assessment.
That’s the promise anyway.
Samsung also mentioned partnerships with healthcare providers. So this isn’t just about consumer gadgets. They’re positioning this as part of a broader healthcare ecosystem.
The Bottom Line
Samsung has built two AI foundation models that learn from wearable biosignal data. xMAE connects PPG and ECG signals for cardiovascular analysis. HiMAE analyzes health patterns across different time scales.
The models run on-device, process data in milliseconds, and don’t need cloud connectivity. They were trained on nearly 10,000 hours of heart data and outperformed existing methods in most evaluation tasks.
Samsung is betting that AI-powered health analysis on wearables is the future. And they’re putting the research behind it.
One Comment