ICML Discuss
Which Statistical Estimators Have Differentially Private Approximations?
by Kamalika Chaudhuri, Daniel Hsu at ICML 2012
Differential privacy is a cryptographically-motivated privacy definition which has gained significant attention over the past few years. Differentially private solutions enforce privacy by adding random noise to the data or a function computed on the data, and the challenge in designing such algorithms is to optimize the privacy-accuracy-sample size tradeoff. This work studies differentially-private statistical estimation, and shows upper and lower bounds on the convergence rates of differentially private approximations to statistical estimators. Our results reveal a connection between differential privacy and the notion of B-robustness in robust statistics, by showing that unless an estimator is B-robust, we cannot approximate it well with differential privacy over a large class of distributions. We then provide an upper bound on the convergence rate of a differentially private approximation to a B-robust estimator with a bounded range. We show that the bounded range condition is necessary if we wish to ensure a strict form of differential privacy.

Related Material

Download PDF Watch Video

Discussion

Email notifications of comments are sent to authors.
Please use the feedback page to report broken links and other problems.
blog comments powered by Disqus