Google's SensorFM Reveals Where AI Takes Wearable Health

Google Research introduced SensorFM, a wearable foundation model pre-trained on over 1 trillion minutes of sensor data from 5 million people. The model uses a ViT-1D encoder with masked-autoencoder objective, ingesting 34 features from five sensors. It outperforms baselines on 34 of 35 health tasks, and its predictions were statistically indistinguishable from clinical ground truth when evaluated by physicians.