Google releases SensorFM, a wearable health model that beats baselines in 34 of 35 tasks

Google researchers have launched SensorFM, a wearable health foundation model. It's trained on data from 5 million people across over 100 countries and more than 20 Fitbit and Pixel Watch models, using over 2 billion hours of signals. The model takes in data from 5 sensors like PPG and accelerometer to track heart rate, sleep, steps, and more.

SensorFM comes in four sizes. The biggest version, SensorFM-B, wins in 33 of 35 health tasks, and its linear probe beats supervised baselines in 34 tasks, covering heart health, metabolism, mental health, and sleep. The team also built an AI agent system that outperformed linear probes in 16 of 20 classification tasks.