Thanks for linking this. I always did wonder how they did it. From a cursory read, it looks like the basic approach is to collect accelerometer data and train it to predict ground truth labels generated via a consumer EEG device. Not sure if each epochs (30s) were considered iids. I'd assume that the overall sequence of activity would be much more informative than a single 30s activity considered independently.
back
1 comments
Sometimes at work I’m concentrating so hard my Apple Watch logs a sleep session…
Definitely not 100% accurate but as someone who checks their sleep log (and time in each stage using the Pillow app) daily, it seems pretty accurate and I can correlate how I feel on a given day back to that day’s sleep report with pretty good correlation.
Definitely not perfect but better than nothing.
If you’ve ever been on medication that suppresses REM sleep, Apple Watch detects REM rebound very clearly.
I won't be surprised. The accelerometers seem to be pretty sensitive. Enough to measure any variations in heart rates and breathing patterns which are likely highly correlated with sleep patterns.
I didn't think it tracked sleep unless you put it into Sleep mode explicitly.
Yeah likewise. If I fall asleep during normal wake hours, such as an unexpected nap, it will vibrate and remind me to stand up because I’ve been inactive for too long.
There should be a toggle for this.
There is. But it’s easy to forget because it’s an unexpected nap