Google's SensorFM Trained on 1 Trillion Minutes of Wearable Data from 5M Users

Google’s SensorFM has been trained on data from five million users, utilizing a total of 1 trillion minutes of wearable data from Fitbit and Pixel Watch devices. The model has outperformed existing benchmarks in 34 out of 35 health and behavioral tasks. While SensorFM could potentially power Google’s AI health coach, the company has not announced any integration plans at this time. This follows Google’s ongoing efforts to leverage large-scale health data for improved AI applications, as seen in previous developments like the ROSA-TFormer model for environmental classification. For practitioners, the performance of SensorFM indicates a significant advancement in processing and interpreting complex health data, which could enhance future health-related AI applications. More details are available in The Decoder.