Title: [Title removed for Anonymity]
Speaker: Graham Cormode, AT & T
Date: Monday, November 29, 2010 12:00 - 1:00 pm
Location: DIMACS Center, CoRE Bldg, Room 431, Rutgers University, Busch Campus, Piscataway, NJ
Differential privacy is a powerful privacy paradigm. The differentially private output of a (randomized) algorithm guarantees that the outputs on two "close" datasets (one including an individual and another excluding an individual) are hard to distinguish. The theory surrounding differential privacy has arisen in recent years, but there are still challenges in applying it.
In this talk, I'll talk about two aspects of applying differential privacy:
(1) Applying privacy when the data is drawn from a high dimensional space and direct application of standard approaches produce vast amounts of noisy output. How can we make privacy scale to high dimensions?
(2) Understanding the limits of the definition: when does differentially private output still reveal information about individuals?
Slides: [Title removed for Anonymity]