Relations among different privacy notions

TL;DR

This paper explores and formalizes the relationships among various privacy notions like differential privacy, Bayesian differential privacy, semantic privacy, and membership privacy, providing theoretical equivalences and implications.

Contribution

It extends semantic privacy to Bayesian semantic privacy, establishes their equivalence with Bayesian differential privacy, and clarifies the relations among multiple privacy notions.

Findings

Bayesian semantic privacy is equivalent to Bayesian differential privacy.

Relations between privacy notions are formally characterized with implications and equivalences.

Improves understanding of how different privacy guarantees relate and can be derived from each other.

Abstract

We present a comprehensive view of the relations among several privacy notions: differential privacy (DP) [1], Bayesian differential privacy (BDP) [2], semantic privacy (SP) [3], and membership privacy (MP) [4]. The results are organized into two parts. In part one, we extend the notion of semantic privacy (SP) to Bayesian semantic privacy (BSP) and show its essential equivalence with Bayesian differential privacy (BDP) in the quantitative sense. We prove the relations between BDP, BSP, and SP as follows: $\epsilon$-BDP $\Longleftarrow$ $\big(\frac{1}{2}-\frac{1}{e^{\epsilon}+1}\big)$-BSP, and $\epsilon$-BDP $\Longrightarrow$ $(e^{2\epsilon}-1)$-BSP $\Longrightarrow$ $(e^{2\epsilon}-1)$-SP. In addition, we obtain a minor result $\epsilon$-DP $\Longleftarrow$ $\big(\frac{1}{2}-\frac{1}{e^{\epsilon}+1}\big)$-SP, which improves the result of Kasiviswanathan and Smith [3] stating $\epsilon$-DP $\Longleftarrow$ $\epsilon/6$-SP for $\epsilon \leq 1.35$. In part two, we establish the relations between BDP and MP. First, $\epsilon$-BDP $\Longrightarrow$ $\epsilon$-MP. Second, for a family of distributions that are downward scalable in the sense of Li et al. [4], it is shown that $\epsilon$-BDP $\Longleftarrow$ $\epsilon$-MP.