Papers › Dirichlet Mechanism for Differentially Private KL Divergence Minimization

Dirichlet Mechanism for Differentially Private KL Divergence Minimization

3 Oct 2021NeurIPS 2021 12arXiv:2110.01984archive 2025-07-28

Donlapark Ponnoprat

Given an empirical distribution f(x) of sensitive data x, we consider the task of minimizing F(y) = D_(KL) (f(x)‖y) over a probability simplex, while protecting the privacy of x. We observe that, if we take the exponential mechanism and use the KL divergence as the loss function, then the resulting algorithm is the Dirichlet mechanism that outputs a single draw from a Dirichlet distribution. Motivated by this, we propose a R\'enyi differentially private (RDP) algorithm that employs the Dirichlet mechanism to solve the KL divergence minimization task. In addition, given f(x) as above and ŷ an output of the Dirichlet mechanism, we prove a probability tail bound on D_(KL) (f(x)‖ŷ), which is then used to derive a lower bound for the sample complexity of our RDP algorithm. Experiments on real-world datasets demonstrate advantages of our algorithm over Gaussian and Laplace mechanisms in supervised classification and maximum likelihood estimation.

PaperPDFConference PDFCode

Code

dirsampling/privatedps officialmentioned in paper report
donlapark/dirichlet-mechanism officialmentioned in paperMIT report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Privacy Preserving

Results from the paper archive 2025-07-28

No leaderboard rows for this paper in the archive.

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections