Coaching and analysis
We leverage a dataset with 40 million hours of wearable knowledge sampled from over 60,000 individuals in the course of the interval from March to Might 2024. The dataset was completely anonymized or de-identified to make sure that participant info was eliminated and privateness was maintained. Topics wore a wide range of Fitbit and Google Pixel smartwatches and trackers and consented for his or her knowledge for use for analysis and growth of recent well being and wellness services. The themes have been requested to self-report intercourse, age, and weight.
To pre-train LSM-2, we make use of the AIM SSL method launched within the earlier part. AIM implements a masked reconstruction coaching goal, and learns to know knowledge that’s naturally lacking, and impute knowledge that’s artificially masked. This unified framework permits LSM-2 to study the underlying construction (together with missingness) inherent in wearable sensor knowledge.
We curate a set of downstream duties to guage the pre-trained mannequin, utilizing meta-data that was collected alongside the sensor alerts for the needs of analysis and growth. These embrace person annotated actions from a set of 20 totally different classes (corresponding to working, snowboarding, kayaking and enjoying golf) and self-reported diagnoses of hypertension and nervousness. These knowledge have been break up into fine-tuning and analysis units the place knowledge from every particular person was solely in both the tuning or the analysis set and never each. Knowledge from people used within the pretraining stage was additionally not included within the fine-tuning or analysis phases.
The generative capabilities of LSM-2 are evaluated via the duties of random imputation, temporal interpolation, temporal extrapolation (forecasting), and sensor imputation, described in our LSM-1 work.
The utility of the LSM-2 embeddings are evaluated by way of linear probe on a variety of discriminative duties. Particularly we gauge the applicability of the LSM-2 embeddings to the duties of binary hypertension classification, binary nervousness classification, and 20-class exercise recognition. We consider LSM-2’s capacity to mannequin physiology by way of age and BMI regression duties.
