English

Approximate Differential Privacy of the $\ell_2$ Mechanism

Cryptography and Security 2025-02-25 v1

Abstract

We study the 2\ell_2 mechanism for computing a dd-dimensional statistic with bounded 2\ell_2 sensitivity under approximate differential privacy. Across a range of privacy parameters, we find that the 2\ell_2 mechanism obtains lower error than the Laplace and Gaussian mechanisms, matching the former at d=1d=1 and approaching the latter as dd \to \infty.

Keywords

Cite

@article{arxiv.2502.15929,
  title  = {Approximate Differential Privacy of the $\ell_2$ Mechanism},
  author = {Matthew Joseph and Alex Kulesza and Alexander Yu},
  journal= {arXiv preprint arXiv:2502.15929},
  year   = {2025}
}