Differential privacy for dummies(github.com)
github.com
Differential privacy for dummies
https://github.com/frankmcsherry/blog/blob/master/posts/2016-02-03.md
6 comments
> The formal version of differential privacy requires that the probability a computation produces any given output changes by at most a multiplicative factor when you add or remove one record from the input.
What is the significance of this? This statement means nothing. The context should be better described. I thought this was a "for dummies" article.
What is the significance of this? This statement means nothing. The context should be better described. I thought this was a "for dummies" article.
I also turned to Wikipedia immediately. Differential privacy is concerned with protecting individual records when making statistical queries. For example, if you could issue a query to a medical case database like »What proportion of the population was born on this specific date in that specific small town with some specific eye color and has this illness?« you could potentially figure out whether or not one person has a specific illness by either getting zero or a very small percentage. One possible solution and what is mostly discussed in the article is adding some noise to the result, as little as possible to not make the query result meaningless but enough to protect each record.
Here is the basic idea of differential privacy in its simplest form.
Suppose that A is a eps-DP algorithm and X1, X2 are two datasets differing by a single point. Recall that for small eps, exp(eps) is roughly (1+eps) and exp(-eps) is roughly (1-eps), so eps-DP yields the guarantee that (roughly speaking), for all subsets S,
Suppose that A is a eps-DP algorithm and X1, X2 are two datasets differing by a single point. Recall that for small eps, exp(eps) is roughly (1+eps) and exp(-eps) is roughly (1-eps), so eps-DP yields the guarantee that (roughly speaking), for all subsets S,
(1-eps) * Pr{ A(X2) \in S } <= Pr{ A(X1) \in S } <= (1+eps) * Pr{ A(X2) \in S }.
The "dummies" are the authors of "Fool's Gold," to which this article is a rebuttal.
https://www.youtube.com/watch?v=5muBxGQ9B2g is a nice introduction and overview
It does a very good job of explaining the basics.
They also published a follow up with the quite popular taxi dataset: http://research.neustar.biz/2014/09/15/riding-with-the-stars...