The Target-Charging Technique for Privacy Accounting across Interactive Computations

TL;DR

The paper introduces the Target Charging Technique (TCT), a novel privacy analysis framework that improves privacy guarantees in interactive differentially private computations by selectively charging only certain computations, thus reducing cumulative privacy loss.

Contribution

The paper presents TCT, a unified framework that generalizes existing privacy tools and enhances privacy analysis for interactive differentially private algorithms.

Findings

TCT allows many computations to be essentially free, reducing overall privacy cost.

TCT generalizes tools like sparse vector and top-k selection for privacy.

TCT extends privacy benefits from Lipschitz functions to general algorithms.

Abstract

We propose the \emph{Target Charging Technique} (TCT), a unified privacy analysis framework for interactive settings where a sensitive dataset is accessed multiple times using differentially private algorithms. Unlike traditional composition, where privacy guarantees deteriorate quickly with the number of accesses, TCT allows computations that don't hit a specified \emph{target}, often the vast majority, to be essentially free (while incurring instead a small overhead on those that do hit their targets). TCT generalizes tools such as the sparse vector technique and top-$k$ selection from private candidates and extends their remarkable privacy enhancement benefits from noisy Lipschitz functions to general private algorithms.