Get Started
Home
Topics
Search
Library
Research questionHow can wearable foundation models organize representations for targeted longitudinal women’s-health prediction?General-purpose wearable models learn broad sensor patterns, but those representations may not align with women’s-health outcomes. This makes it difficult to transfer a shared longitudinal representation across related tasks without losing task-specific predictive performance.
AI
Evaluation & Benchmarks
Health
Machine Learning
Multimodal Models
Research Paper
Technology
Latest papersRecent research connected to this question, newest first.FemWear: A Parameter-Efficient Wearable Foundation Model for Women's HealthThe source studies FemWear, which repurposes a pretrained multimodal wearable backbone for menstrual, symptom, affective, sleep/recovery, autonomic, activity, and pregnancy outcomes using frozen patch projection and Transformer components with parameter-efficient adapters and task-family heads. Evidence comes from six cohorts, fixed participant splits, OpenMHC retention tests, and a stricter 42-participant nested leave-one-participant-out audit; improvements were mixed under the stricter audit, with no endpoint showing a strictly positive corrected confidence interval. Capacity-matched results exceeded a latest-day multilayer perceptron but not shared-GRU or mixture-of-experts baselines.research paper · Sep 2, 2026
Related questions
How can language models reason over noisy, longitudinal wearable records while integrating multiple physiological signals?How can wearable-derived mobility outcomes be modelled longitudinally across diseases without paired participants?How can synthetic medical time-series generation preserve rare-class patterns across temporal scales for downstream prediction?How should tabular foundation models be adapted for censored time-to-event prediction under competing risks?