Researchfoundation modeltime serieswearableshealthcare
JETS Predicts Medical Conditions From Wearables
8.0
Researchers at MIT and Empirical Health developed JETS, a self-supervised joint-embedding time-series foundation model trained on roughly 3 million person-days of Apple Watch data, and reported the work in a paper accepted to a NeurIPS workshop. JETS leverages masking and embedding prediction to learn from 85% unlabeled, irregular wearable records and achieved AUROCs such as 86.8% for high blood pressure and 81% for chronic fatigue syndrome.
Key Points
- 1Develops JETS, a self-supervised JEPA-based time-series foundation model trained on ~3 million person-days.
- 2Demonstrates strong discrimination (AUROC up to 86.8%) for multiple cardiac and behavioral conditions.
- 3Enables use of irregular, unlabeled wearable data via pretraining, improving downstream clinical prediction on scarce labels.
Scoring Rationale
Strong empirical validation and workshop acceptance, but mainly an architecture adaptation focused on one vertical limits broader novelty.
Sources
Public references used for this report.
Practice with real Health & Insurance data
90 SQL & Python problems · 15 industry datasets
Used by DS/ML engineers at top companies
250 free problems · No credit card
See all Health & Insurance problems
