Connect the Dots: Tighter Discrete Approximations of Privacy Loss Distributions

Abstract

The privacy loss distribution (PLD) provides a tight characterization of the privacy loss of a mechanism in the context of differential privacy (DP). Recent work [18–20, 24] has shown that PLD-based accounting allows for tighter (ε, δ)-DP guarantees for many popular mechanisms compared to other known methods. A key question in PLD-based accounting is how to approximate any (potentially continuous) PLD with a PLD over any specified discrete support. We present a novel approach to this problem. Our approach supports both pessimistic estimation, which overestimates the hockey-stick divergence (i.e., δ) for any value of ε, and optimistic estimation, which underestimates the hockey-stick divergence. Moreover, we show that our pessimistic estimate is the best possible among all pessimistic estimates. Experimental evaluation shows that our approach can work with much larger discretization intervals while keeping a similar error bound compared to previous approaches and yet give a better approximation than an existing method [24].

High-Fidelity Synthetic Data Applications for Data Augmentation

“…Advanced privacy accounting [54] involves systematically measuring and evaluating the privacy guarantees provided by synthetic data generation methods. This technique assesses the level of privacy protection offered by synthetic datasets and quantifies the potential risks of re-identification or privacy breaches.…”

Unified Interpretation of the Gaussian Mechanism for Differential Privacy Through the Sensitivity Index

“…To further examine the effect of mechanism parameter choices on the ∆-divergence, Figure 6 investigates switching from a base SGM M with p " 0.01, N " 500 and σ " 0.54 to Ă M, where r p P r0.04, 0.9s, r N P r534, 1500s and the resulting r σ P r0.55, 21s. All mechanisms are calibrated to p8, 10 ´5q-DP using the numerical system by Doroshenko et al (2022) and the absolute calibration error in terms of ε is ď 0.00042. A monotonic increase in the ∆-divergence with the noise multiplier is observed, culminating in a maximum divergence value of around 0.12.…”

A Petri nets-based approach to modeling SCORM sequence

“…First, repeatedly accessing sensitive data leads to an accumulation of privacy loss. Such a privacy loss can be bounded by the sequential composition theorem or higher-order techniques (e.g., moments accountant [1,12]). Second, DP is not affected by post-processing.…”