Gaussian Differential Privacy

Follow this topic Ask about this topic

Gaussian differential privacy represents privacy with a parameter μ tied to the trade-off between correctly detecting whether a person’s data was used and making false alarms. Researchers derive and compose these trade-off functions, including for subsampled DP-SGD, and use tools such as privacy-loss distributions and f-DP accounting to analyze complex pipelines. Applications include private model training and synthetic data, where tighter accounting can reduce the noise needed for a given privacy level. A key challenge is aligning theoretical guarantees and conversions between privacy measures with empirical audits of real mechanisms and their practical privacy-utility costs.

Latest papers

11 papers4 papers in the last four weeks, against 2 the four weeks before. 0.0% of all new papers.

Gaussian Approximation for Composition and Subsampling of Differentially Private Algorithms
Oct 23, 2026Vitaly Feldman, Moshe ShenfeldDifferential Privacy Gaussian Differential Privacy
Apple · The Hebrew University of Jerusalem
Privacy Accounting for Empirical Denoisers
Oct 20, 2026William Cheng, Dongjun Kim, Mijung ParkDifferential Privacy Diffusion Sampling
University of British Columbia · Stanford University · Now at Adobe.
Tight End-to-End Accounting for Differentially Private Synthetic Data
Oct 12, 2026Georgi Ganev, Vadym Doroshenko, Arun Ganesh +3Differential Privacy Differentially Private Synthetic Data Generation
UCL · SAS · Google +3
Certification-Based Differentially Private Learning
Sep 30, 2026Mihnea Ghitu, Matthew WickerAbstract Interpretation Privacy-Preserving ML
Department of Computing Imperial College London London, United Kingdom
Empirical Auditing of Edge-Private Graph Generators
Sep 21, 2026Anum Fatima, Stratis Limnios, James Adams +4Privacy Auditing Edge Inference
Department of Statistics University of Oxford Oxford, United Kingdom · Plenitude Consulting, London, United Kingdom · National Physical Laboratory Teddington, United Kingdom +3
Component-Aware Differential Privacy for Federated Multilingual Speech-LLMs
Sep 10, 2026Jordi Luque, Fernando López, Aleix SantPrivacy-Preserving Language Models Differentially Private Federated Learning
Scientific Research Telef´onica Innovaci´on Digital Barcelona, Spain · Telef´onica Innovaci´on Digital Universidad Aut´onoma de Madrid Madrid, Spain · Scientific Research Telef´onica Innovaci´on Digital Universitat Polit`ecnica de Catalunya Barcelona, Spain
EdgeRefine: Privacy-Utility Balance for Graphs via Jaccard Sampling under Edge Differential Privacy
Jul 9, 2026Wenxiu Ding, Muzhi Liu, Zheng Yan +3Graph Neural Networks Privacy-Preserving ML
State Key Laboratory of Integrated Services Networks, School of Cyber Engineering, Xidian University, Xi’an, Shaanxi, China · School of Cyber Engineering, Xidian University, Xi’an, Shaanxi, China
On Choosing the $μ$ Parameter in Gaussian Differential Privacy
Jun 8, 2026Bogdan Kulynych, Antti HonkelaDifferential Privacy Gaussian Differential Privacy
Biomedical Data Science Center, Lausanne University Hospital · University of Helsinki
Differentially Private Motif-Preserving Multi-modal Hashing
May 14, 2026Zehua Cheng, Wei Dai, Jiahao SunCross-Modal Learning Cross-Modal Retrieval
Department of Computer Science University of Oxford Oxford, United Kingdom · FLock.io London, United Kingdom
Trade-off Functions for DP-SGD with Subsampling based on Random Shuffling: Tight Upper and Lower Bounds
May 7, 2026Marten van Dijk, Murat Bilgehan ErtanDifferentially Private Stochastic Gradient Descent Differential Privacy
CWI Amsterdam · ∗Affiliated with Vrije Universiteit Amsterdam.
Tight Auditing of Differential Privacy in MST and AIM
Apr 20, 2026Georgi Ganev, Meenatchi Sundaram Muthu Selva Annamalai, Bogdan KulynychPrivacy Auditing Synthetic Data Generation
1UCL · 2SAS · 3Lausanne University Hospital

Oct 23, 2026J/K move · Enter open · S save

Gaussian Approximation for Composition and Subsampling of Differentially Private Algorithms

Vitaly Feldman, Moshe Shenfeld

Apple · The Hebrew University of Jerusalem

Differential Privacy Gaussian Differential Privacy Differentially Private Stochastic Gradient Descent

Open paper AskPDF arXiv

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 of notions and concentration results. Yet, existing closed-form bounds either lose a logarithmic factor or require restrictive conditions on the noise scale and delicate analysis of subsampling. Further, while it is well known that sums of independent random variables are approximately Gaussian, existing CLT-based techniques give an approximation with a large error in

δ

. We show that the privacy loss of this pair is approximately Gaussian as long as each step has a small privacy loss except with small probability. Specifically, its privacy loss distribution is close to a Gaussian with the same mean and variance, after truncating the rare large privacy losses. The closeness in both directions is up to a shift of

\varepsilon

that grows only as

\ln(1/β)

for an additive error

β

δ

and is a lower-order term in typical applications. In particular, the composition satisfies Gaussian differential privacy up to a lower-order shift of

\varepsilon

and the additive error

β

. We also give a simple way to analyze the resulting variance-based bounds for subsampled mechanisms which relies on the

χ^2

divergence of a single step. For the Gaussian mechanism at sampling rate

1/t

and

t>\ln^{3}(1/δ)

, this gives the predicted rate for

σ\gtrsim\max\{1,\sqrt{\ln(1/δ)}/\ln(t)\}

, whereas the standard Rényi-divergence analyses need

σ\gtrsim\sqrt{t\ln(1/δ)}/\ln(t)

. We extend these results to random allocation (balls-and-bins sampling), and show that it explains the high accuracy of the connect-the-dots method.