Skip to content
Preprint

Diffusion Sensitivity Geometry and Strong Duality for Optimal Privacy-Utility Tradeoffs

Oct 2026 · 0 citations · 34 references
Computer Science Mathematics

TL;DR

A framework for the optimal privacy-utility tradeoff of ideal diffusion denoisers is developed, replacing conservative bounds with exact sensitivity characterizations and provably optimal noise calibration, and exact privacy guarantees for R\'enyi, approximate, and Gaussian differential privacy are obtained.

Abstract

Existing approaches to differential privacy in diffusion models rely on loose sensitivity bounds and isotropic Gaussian noise, leading to unnecessary estimation error. We develop a framework for the optimal privacy-utility tradeoff of ideal diffusion denoisers, replacing conservative bounds with exact sensitivity characterizations and provably optimal noise calibration. First, we show that worst-case sensitivity over datasets of arbitrary size can be characterized exactly using only two records. For a data domain of size $m$, this reduces the analysis to at most $m(m-1)$ constraints, regardless of dataset size. Our characterization can be arbitrarily tighter than conventional diameter-based sensitivity bounds, revealing potentially unbounded improvements in noise calibration. Second, we obtain exact privacy guarantees for R\'enyi, approximate, and Gaussian differential privacy, as well as mutual-information privacy for binary records, avoiding the looseness of previous generic privacy conversions. Finally, we show that the minimum achievable weighted mean-squared error can be computed through a semidefinite program and prove strong duality, providing certificates of global optimality.

View source

Similar papers

Preprint Oct 2026

Gaussian Approximation for Composition and Subsampling of Differentially Private Algorithms

The privacy analysis of differentially private stochastic gradient descent (DP-SGD) relies on bounds for the composition of subsampled mechanisms. The privacy of such a composition is bounded by that of a pair of product distributions, whose privacy loss is a sum of independent terms and has been analyzed via a variety...

V. Feldman, M. Shenfeld · 2 citations
Preprint Oct 2026

Exact Privacy--Distortion Design of Support-Constrained Randomized Response

Under approximate local privacy restricted to input pairs at distance at most $H$, it is proved that stringent approximate privacy forces the support size to scale as $\Omega(H/\delta)$.

A. Zamani, Parastoo Sadeghi, M. Skoglund · 0 citations
Review Oct 2026

Learning Sparse Support under Differential Privacy: Adaptive Algorithms and Minimax Limits

Saturated Propose-test-release, a general mechanism that privately releases the output of a discrete selector with probability one once its stability certificate reaches a finite threshold, is introduced and a computable support-stability score is constructed.

Jian-Zhong Gu, T. Cai · 1 citation
Open access Sep 2026

An (ε, δ)-Differential Privacy Analysis of Adaptive Noise Calibration via Hockey-Stick Divergence

A two-stage adaptive mechanism that first obtains a privacy-preserving estimate of how spread out the data are, using the interquartile range (IQR) as the measure of dispersion, and then uses that estimate to shrink the noise added to the query answer whenever the data are concentrated.

Bekir Danış, Muhammed Hasdemir · 0 citations
#machine learning Preprint Oct 2026

On Inherent Privacy of Posterior Sampling: A Unified R\'enyi-Divergence Framework

It is shown that the privacy loss incurred by One-Posterior-Sampling (OPS) is governed by posterior exponential moments of the single-record log-likelihood ratio, and explicit DP guarantees under uniformly bounded, sub-Gaussian, or sub-exponential tail regimes are established.

Ru-Yu Zhou, Fang Liu · 0 citations
Preprint Nov 2026

Differentially Private Distributed Optimization via the Intrinsic Randomness of Zeroth-Order Estimators: Sampling Geometry, Convergence, and Privacy Guarantees

Distributed zeroth-order optimization has gained increasing attention in applications where exact gradients are computationally expensive or unavailable. However, the iterative exchange of local decision variables makes distributed networks vulnerable to inference attacks. While differential privacy provides a rigorous...

Jia-Jun Wang, Yujie Tang · 0 citations

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.