Google has described SensorFM, a wearable data model trained on de-identified sensor data from five million people who consented to the use of their data for health and wellness research. The data was captured between September 2024 and September 2025 and spans more than 100 countries, all 50 U.S. states, and over 20 Fitbit and Pixel Watch device models.
According to the source, the pre-training corpus contains over two billion hours, or more than a trillion minutes, of minute-resolution signals. For each person, the dataset includes several weeks of data.
SensorFM ingests 34 one-minute aggregate features derived from five sensor modalities: photoplethysmography (PPG), accelerometry, electrodermal activity (EDA), skin temperature, and altimetry. These signals capture heart rate and heart-rate variability, blood-oxygen saturation, sleep stages, motion and steps, skin conductance, and temperature over a full 24-hour window.
The model is built on self-supervised reconstruction using the LSM-2 approach and its Adaptive and Inherited Masking (AIM) framework. The source says this matters because wearable data is often missing or fragmented when sensors power-cycle, devices come off the wrist, power saving modes are used, or sensors switch on and off.
Instead of imputing gaps or discarding incomplete windows, AIM treats missingness as part of the data and learns directly from incomplete recordings. It combines tokens inherited from genuine gaps with those artificially masked for the reconstruction objective and treats the two as equivalent.
Google says SensorFM is missingness-aware by construction and that it uses fragmented data productively.
Source: research.google.
Companies can share verified announcements through Newz9’s international press release submission page.
