The goal is to be able to identify you, not recover an image of you.

This is just as much PII as your fingerprints. Fingerprint based devices store metadata not images. I can't look at a fingerprint and connect it to a person, but a machine can. Similarly the output of a NN designed to recognize people will be useful for validation. Otherwise biometrics would not work.

Remember if they are using your photo as training data the NN is going to learn you vs someone else or it's useless. If they can find you in any photo that's as effective as handing a person the uploaded photo and giving the same task.

I understood your question. What I am pointing out is that when you submit an artifact to FB, it is a trivial matter for them to store as 'extended-attributes' your facial biometrics. This has nothing to do with the ML pipeline that munches on data in the aggregate.

As to whether it is possible to extract such information, per my understanding the internals of ML pipelines are rather opaque, so it would be non-trivial, if not impossible. I have no idea either way.

I think there are two separate questions being raised. My understanding (EDIT: my understanding of the concern being raised in this discussion) was not that they would try to recover individual readings from a trained model, but that they would run a model on your face to extract features, then save an array of features that defines your face. Only those features are needed to identify your face - after data extraction the photo is no longer needed.

This article talks about how those data points are extracted using a different method.

https://medium.com/@ageitgey/machine-learning-is-fun-part-4-...

It is not about being able to recover, rather about being able to identify your face on other photos.