I worked on one of the first wearable foundation models in 2018. The innovation of this 2025 paper from Apple is moving up to a higher level of abstraction: instead of training on raw sensor data (PPG, accelerometer), it trains on a timeseries of behavioral biomarkers derived from that data (e.g., HRV, resting heart rate, and so on.).
They find high accuracy in detecting many conditions: diabetes (83%), heart failure (90%), sleep apnea (85%), etc.
crorella 1 minutes ago [-]
Insurance and health insurance companies must be super interested in this research and its applications.
throwaway314155 1 hours ago [-]
Had the phrase "foundation model" become a term of art yet?
brandonb 31 minutes ago [-]
By 2018, the concept was definitely in the air since you had GPT-1 (2018) and BERT (2018). You could argue even Word2Vec (2013) had the core concept of pre-training on an unsupervised or self-supervised objective leading to performance on a downstream semantic task. However, the phrase "foundation model" wasn't coined until 2021, to my knowledge.
puppymaster 3 hours ago [-]
reminds me of Jim Simons of Renaissance advise when it comes to data science - sort first, then regress.
Not every day you find pseudo permutation in the wild!
clickety_clack 1 hours ago [-]
The guy was sorting the X separately from y? That can’t be a real.
1 hours ago [-]
LPisGood 54 minutes 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.
aanet 17 minutes ago [-]
Thanks for posting this. This looks promising...
I have about 3-3.5 years worth of Apple Health + Fitness data (via my Apple Watch) encompassing daily walks / workouts / runs / HIIT / weight + BMI / etc. I started collecting this religiously during pandemic.
The exported Fitness data is ~3.5GB
I'm looking to do some longitudinal analysis - for my own purposes first, to see how certain indicators have evolved.
Has anyone done something similar? Perhaps in R, Python? Would love to do some tinkering. Any pointers appreciated!
Thanks!!
kridsdale1 14 minutes ago [-]
It might actually be worth writing your analysis in Swift with the actual HealthKit API and visualization libraries.
Bonus: when you’re done, you’ll have an app you can sell.
vibecodermcswag 3 hours ago [-]
i love this because I build in medtech, but the big problem is no open weights, nor open data.
you can export your own apple XML data for usage and processing, but if you want to create an application and request apple XML data from users, that likely crosses into clinical research territory with data security policy requirements and de-identification needs.
pricklyprice 2 hours ago [-]
what is the best way for non-big tech to buy such data for research and product development?
guzik 1 hours ago [-]
Some are for free:
- aidlab.com/datasets
- physionet.org
piratesAndSons 56 minutes ago [-]
Trusting your health data with AI brothers is... extremely ill-advised.
I don't even trust Apple themselves, which will sell your health data any insurance company any minute now.
kridsdale1 13 minutes ago [-]
What do you base that suspicion on?
fiduciarytemp 3 hours ago [-]
Has anyone seen the publishing of the weights or even an API release?
brandonb 3 hours ago [-]
In the paper, they say they can't release the weights due to terms of consent with study participants (this is from the Apple Heart and Movement study).
memming 1 hours ago [-]
Interesting to see contrastive loss instead of a reconstruction loss.
dyauspitr 3 hours ago [-]
Is there a way to run this on your own data? I’ve been wearing my Apple Watch for years and would love to be able to use it better.
brandonb 3 hours ago [-]
Not yet -- this one is just a research study. Some of their previous research has made it into product features.
Apple was reporting VO2max for a very long time (much before 2023). I wonder what the accuracy was back then? Maybe they should the option for users to re-compute those past numbers based on the latest and greatest algorithm.
MangoToupe 1 hours ago [-]
Can someone explain what "wearable foundation" means?
compiler-guy 1 hours ago [-]
It's a "Foundation Model" for wearable devices. So "wearable" describes where it is to be used, rather than describing "foundation".
They find high accuracy in detecting many conditions: diabetes (83%), heart failure (90%), sleep apnea (85%), etc.
https://stats.stackexchange.com/questions/185507/what-happen...
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 have about 3-3.5 years worth of Apple Health + Fitness data (via my Apple Watch) encompassing daily walks / workouts / runs / HIIT / weight + BMI / etc. I started collecting this religiously during pandemic.
The exported Fitness data is ~3.5GB
I'm looking to do some longitudinal analysis - for my own purposes first, to see how certain indicators have evolved.
Has anyone done something similar? Perhaps in R, Python? Would love to do some tinkering. Any pointers appreciated!
Thanks!!
Bonus: when you’re done, you’ll have an app you can sell.
you can export your own apple XML data for usage and processing, but if you want to create an application and request apple XML data from users, that likely crosses into clinical research territory with data security policy requirements and de-identification needs.
- aidlab.com/datasets
- physionet.org
I don't even trust Apple themselves, which will sell your health data any insurance company any minute now.
For example, Apple Watch VO2Max (cardio fitness) is based on a deep neural network published in 2023: https://www.empirical.health/blog/how-apple-watch-cardio-fit...