3 ms·
I'm not sure I want to open this can of worm here, but this is one of the appealing properties of differential privacy: there is a quantitative bound you can in
by frankmcsherry 9y ago
I'm not sure I want to open this can of worm here, but this is one of the appealing properties of differential privacy: there is a quantitative bound you can invoke ("we will mine at most eps-dp from your emails, in exchange for this tasty cookie") and then you don't have to have a complicated discussion about which types of PII they ripped from your email, metadata vs non-, and other increasingly meaningless distinctions.
Differential privacy also has the nice property that it works like a fungible currency: it is additive, so even if you hand out x, y, and z units to different folks, they collectively can't spend more than x + y + z units, no matter how clever they are. In principle this could help folks determine a value for their sensitive data, and perhaps set up decent standards for people to value their own privacy (e.g. "selling 0.1 units of differential privacy from your email could shift your health insurance premiums by 10%").
- TeMPOraL 9y agoIn what state is the field of this differential privacy right now? Is anything useful already deployed / expected to be deployed soon, or does it require much more research?
- frankmcsherry 9y agoGoogle uses it.[0] I think it mostly needs interested parties; folks who care enough to commit to doing something more interesting with it than just counting things. I've certainly built several tools while at MSFT, they just .. languished. [0]: https://github.com/google/rappor https://github.com/google/rappor
- physicsyogi 9y agoApple uses it[0] and lets you opt out of sharing your data as well[1]. [0]: https://www.wired.com/2016/06/apples-differential-privacy-collecting-data/ https://www.wired.com/2016/06/apples-differential-privacy-co... [1]: https://www.howtogeek.com/290298/what-is-differential-privacy-and-how-does-it-keep-my-data-anonymous/ https://www.howtogeek.com/290298/what-is-differential-privac...