3 ms·
This wouldn't be such an issue if they couldn't see the data on which they are computing (and individual data wouldn't be exposed in data breaches either). But
by mtgx 9y ago
This wouldn't be such an issue if they couldn't see the data on which they are computing (and individual data wouldn't be exposed in data breaches either).
But I'm not sure if Google even cares too much about doing that (I know they've experimented with this, but nothing on a scale that matters), and Facebook certainly doesn't. Apple seems to be the only one that does somewhat with its differential privacy approach.
I think they can all do much more, but they're not trying too hard.
- deong 9y agoI'm not sure you understand what differential privacy is. Apple still has your data. If there's a breach, an attacker gets your data. It's probably encrypted, just as it is with pretty much any modern system where user data is collected, but differential privacy doesn't have anything to do with what data is collected or stored. Differential privacy is just a method for limiting how much information can be reliably inferred about an entire dataset from limited queries on statistical properties of the data. It's basically this. Imagine I have two data sets that are identical, except one has your specific data and the other doesn't. It's possible in some circumstances to infer information about you by asking for things like averages on the datasets. Differential privacy is a method for defeating attacks that rely on that type of "leakage" of information. It assures that if two datasets are close enough, queries on those datasets will not yield significant information about their differences (i.e., the presence or absence of one particular data point won't be detectable). It's all about how queries of the data perform, and not at all about the data itself.