Comment by LPisGood
2 days ago
Is anyone else surprised by how poorly performing the results are for the vast majority of cases? The foundation model which had access to sensor data and behavioral biomarkers actually _underperformed_ the baseline predictor that just uses nonspecific demographic data in almost 10 areas.
In fact, even when the wearable foundation model was better, it was only marginally better.
I was expecting much more dramatic improvements with such rich data available.
I wonder how much of that is driven by poor performing behavioral models. There was a HN article from a few weeks back and it only had an accuracy of about 70% determining if someone was awake or asleep. I would guess that the secondary behavioral data used in this data (like cardiovascular fitness) are much harder to predict from raw sensor data than being awake or asleep.
I worked with similar data in grad school. I'm not surprised. You can have a lot of data, but sometimes the signal (or signal quality) just isn't present in that haystack, and there's nothing you can do about it.
Sometimes you just have to use ultrasound or MRI or stick a camera in the body, because everything else might as well be reading tea leaves, and people generally demand very high accuracy when it comes to their health.