English

Optimal Noise Adding Mechanisms for Approximate Differential Privacy

Data Structures and Algorithms 2013-12-20 v3 Cryptography and Security

Abstract

We study the (nearly) optimal mechanisms in (ϵ,δ)(\epsilon,\delta)-approximate differential privacy for integer-valued query functions and vector-valued (histogram-like) query functions under a utility-maximization/cost-minimization framework. We characterize the tradeoff between ϵ\epsilon and δ\delta in utility and privacy analysis for histogram-like query functions (1\ell^1 sensitivity), and show that the (ϵ,δ)(\epsilon,\delta)-differential privacy is a framework not much more general than the (ϵ,0)(\epsilon,0)-differential privacy and (0,δ)(0,\delta)-differential privacy in the context of 1\ell^1 and 2\ell^2 cost functions, i.e., minimum expected noise magnitude and noise power. In the same context of 1\ell^1 and 2\ell^2 cost functions, we show the near-optimality of uniform noise mechanism and discrete Laplacian mechanism in the high privacy regime (as (ϵ,δ)(0,0)(\epsilon,\delta) \to (0,0)). We conclude that in (ϵ,δ)(\epsilon,\delta)-differential privacy, the optimal noise magnitude and noise power are Θ(min(1ϵ,1δ))\Theta(\min(\frac{1}{\epsilon},\frac{1}{\delta})) and Θ(min(1ϵ2,1δ2))\Theta(\min(\frac{1}{\epsilon^2},\frac{1}{\delta^2})), respectively, in the high privacy regime.

Keywords

Cite

@article{arxiv.1305.1330,
  title  = {Optimal Noise Adding Mechanisms for Approximate Differential Privacy},
  author = {Quan Geng and Pramod Viswanath},
  journal= {arXiv preprint arXiv:1305.1330},
  year   = {2013}
}

Comments

27 pages, 1 figure

R2 v1 2026-06-22T00:12:24.724Z