RealFeedApp HiPPT aligns recording level with context
Researchers created a corpus of over 4 million pulse recordings from 202k users to train models. The HiPPT method combines individual pulse shape alignment with context alignment of recordings sharing identical metadata.
- Field
- training language models
- What they did
- Developed a method for training models on wearable device data by combining pulse shape alignment with textual context descriptions.
- Why it matters
- This enables models to better understand physiological patterns using data from hundreds of thousands of users.
#ppg#wearable#pretraining#physiology#representation learning
