[Pl-seminar] 4/24: Marco Gaboardi, "A language for Differential Privacy"

Vincent St-Amour stamourv at ccs.neu.edu
Thu Apr 18 13:37:16 EDT 2013

NEU Programming Languages Seminar presents

Marco Gaboardi
University of Pennsylvania

Wednesday, 4/24
Room 366 WVH (http://www.ccs.neu.edu/home/wand/directions.html)

A language for Differential Privacy

Differential privacy offers a way to answer queries about sensitive
information while offering strong, provable privacy guarantees. Several
tools have been developed for certifying that a given query is
differentially private. In this talk, I will present one approach, based
on a functional programming language named Fuzz, that we are developing
at the University of Pennsylvania.

One of the most common mechanisms for turning a (possibly
privacy-leaking) query into a differentially private consists in adding
noise to the result of a query. To ensure the privacy guarantee, the
noise must be proportional to the query sensitivity: how much the result
of a query can change with respect to linear changes in the input. Fuzz
uses linear indexed types and lightweight dependent types to express a
rich sensitivity analysis, and a probability monad to express randomized
computation. The soundness of Fuzz guarantees that any program that has
a certain type is differentially private. So, Fuzz can be used to
certify differentially private a broad class algorithms. Moreover, its
type analysis can be used more in general to perform sensitivity
checking and sensitivity inference.

More information about the pl-seminar mailing list