Limiting Privacy Breaches in Differential Privacy

In recently years, privacy-preserving data mining has become more import and attractedmore attention from data mining community. Among the existing privacy preserving models, E -differential privacy provides the strongest privacy guarantees and has no assumption about the adversary's background information and compute ability. However, howto set E to satisfy privacy is still an open problem. In this paper, we propose a tactic, named LPB (Limiting Privacy Breaches), to set the privacy parameter intuitively. LPB ensures that, if the prior belief about individual is bounded by some threshold, the posterior belief, after given the published randomized result, is no more than another threshold.