Google Research Introduces SensorFM: A Wearable Health Foundation Model Pretrained on One Trillion Minutes of Sensor Data

A foundation model for wearable health was pretrained on over one trillion minutes of sensor data collected from millions of people across multiple countries and device types. Traditional wearable health models are built separately for each specific health outcome, but this approach becomes impractical when dealing with many different prediction tasks. The new model demonstrates that scaling both the model size and training data proportionally improves performance across 35 different health prediction tasks, including cardiovascular, metabolic, mental health, sleep, and lifestyle indicators. The model handles missing sensor data more effectively than conventional methods by learning to reconstruct gaps that occur during normal device use, and its learned representations can be adapted to new health predictions with minimal additional training.