Differentially private techniques guarantee privacy while often permitting accurate answers

...this is something I have been reading up on lately as a means of preserving user privacy/anonymity. As of yet, however, I haven't found any examples of it being applied in a real-world setting. Instead, I am seeing it in academic papers, and discussed mathematically.

Is anyone aware of how exactly this would be applied to real-world data, or indeed, know of any examples? I'm finding it very hard to make the transition from reading about it, to how I might actually go about applying it. If nobody else is applying it then I doubt I have the skills to be the first and so don't want to waste my time flogging a dead horse.

  • I would not trust the phrase "guarantee privacy" in the quote above. The techniques are proven imperfect. See arxiv.org/abs/cs/0610105 for a paper by researchers who broke Netflix anonymity by cross referencing data from IMDB. Commented Jan 7, 2016 at 22:19

3 Answers 3


Here is an article from Scientific American at the end of 2012 mentioning one real-world example-- 'OnTheMap' and references research into others:


"One real-world application already uses differential privacy: a Census Bureau project called OnTheMap, which gives researchers access to agency data. Also, differential privacy researchers have fielded preliminary inquiries from Facebook and the federally funded iDASH center at the University of California, San Diego, whose mandate in large part is to find ways for researchers to share biomedical data without compromising privacy."


Differentially private mechanisms have been implemented in prototypes. These prototypes will usually take a set of records (eg. a database table) and a query, and then return the result of the query on the table with a small amount of added noise to guarantee differential privacy. While proving that a mechanism is differentially private is mathematically demanding, implementing mechanisms is usually straightforward, and involves simply adding noise to the result of the query from some well defined distribution.

The Microsoft Research PINQ project allows you to define queries in a LINQ-like syntax, and will execute them and protect the results using a diferentially private mechanism. The tutorial is here: http://research.microsoft.com/en-us/projects/pinq/tutorial.aspx

The Airavat system uses the Sample-and-aggregate differentially private mechanism to protect map-reduce query results using differential privacy. Prototype code can be found on the bottom of the project page: http://z.cs.utexas.edu/users/osa/airavat/

A couple more systems (with available code) include Fuzz (UPenn) and GUPT (Berkeley).


The biggest deployment of differential privacy, to date (that anyone is talking about publically), is Google's RAPPOR project, which is used to report usage statistics for Google Chrome.

You can read about it here: https://googleonlinesecurity.blogspot.com/2014/10/learning-statistics-with-privacy-aided.html

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .