Randomized Response Schemes, Privacy and Usefulness
Lists
Only you see this
Abstract
We define the notions of weak and strong usefulness and investigate the question whether a randomized response scheme can combine two conflicting goals, namely being weakly (or even strongly) useful and, at the same time, providing ε-differential privacy. We prove the following results. First, if a class F cannot be weakly SQ-learned under the uniform distribution, thenε-differential privacy rules out weak usefulness. This extends a result from [6] that was concerned with the class of parities. Second, for a broad variety of classes F that actually can be weakly SQ-learned under the uniform distribution, we design a randomized response scheme that provides ε-differential privacy and strong usefulness.
BibTeX
@inproceedings{alda2014randomized,
author = {Francesco Aldà and Hans Ulrich Simon},
title = {{Randomized Response Schemes, Privacy and Usefulness}},
booktitle = {AISec@CCS},
year = {2014},
doi = {10.1145/2666652.2666654},
}Loading related content...