We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Data Structures and Algorithms

Title: Tight Bounds for Differentially Private Anonymized Histograms

Abstract: In this note, we consider the problem of differentially privately (DP) computing an anonymized histogram, which is defined as the multiset of counts of the input dataset (without bucket labels). In the low-privacy regime $\epsilon \geq 1$, we give an $\epsilon$-DP algorithm with an expected $\ell_1$-error bound of $O(\sqrt{n} / e^\epsilon)$. In the high-privacy regime $\epsilon < 1$, we give an $\Omega(\sqrt{n \log(1/\epsilon) / \epsilon})$ lower bound on the expected $\ell_1$ error. In both cases, our bounds asymptotically match the previously known lower/upper bounds due to [Suresh, NeurIPS 2019].
Comments: To appear in SOSA 2022
Subjects: Data Structures and Algorithms (cs.DS); Cryptography and Security (cs.CR)
Cite as: arXiv:2111.03257 [cs.DS]
  (or arXiv:2111.03257v1 [cs.DS] for this version)

Submission history

From: Pasin Manurangsi [view email]
[v1] Fri, 5 Nov 2021 04:49:18 GMT (17kb)

Link back to: arXiv, form interface, contact.