Module 5: Future Directions

AI

  • Large opportunity for Self-Supervised Learning
  • Easy to collect, may be hard to characterize

Foundation Models

Inertia-1

  • Inertia-1 - foundation model from NHANES+UKBB raw data
  • https://yang-ai-lab.github.io/Inertia-1/
  • https://arxiv.org/abs/2607.06617
  • https://github.com/yang-ai-lab/Inertia-1#pretraining-data-nhanes

SensorFM

Google using FitBit/Pixel Watch: - https://research.google/blog/sensorfm-towards-a-general-intelligence-and-interface-for-wearable-health-data/

StepFM

  • Step Foundation Model
  • https://arxiv.org/pdf/2607.06954

PAT

  • Pretrained Actigraphy Transformer
    • patch embeddings and masked reconstruction pretraining on large-scale NHANES actigraphy data,
    • evaluated on mental-health-related tasks such as depression, sleep disorders, and medication usage.
  • Based on full 7 days (10080 vector)

Landscape

From the StepFM model paper:

Modules

https://jhuwit.github.io/wearabler/modules

References