Precision-based attacks and interval refining: how to break, then fix, differential privacy on finite computers
Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 kernel pith:OFBV5DRSrecord.jsonopen to challenge →
read the original abstract
Despite being raised as a problem over ten years ago, the imprecision of floating point arithmetic continues to cause privacy failures in the implementations of differentially private noise mechanisms. In this paper, we highlight a new class of vulnerabilities, which we call \emph{precision-based attacks}, and which affect several open source libraries. To address this vulnerability and implement differentially private mechanisms on floating-point space in a safe way, we propose a novel technique, called \emph{interval refining}. This technique has minimal error, provable privacy, and broad applicability. We use interval refining to design and implement a variant of the Laplace mechanism that is equivalent to sampling from the Laplace distribution and rounding to a float. We report on the performance of this approach, and discuss how interval refining can be used to implement other mechanisms safely, including the Gaussian mechanism and the exponential mechanism.
This paper has not been read by Pith yet.
Forward citations
Cited by 6 Pith papers
-
Auditing Apple's DifferentialPrivacy.framework: Implementation Bugs, Misconfigurations, and Practical Risks
Audit finds DP violations in 5 of 9 mechanisms in Apple's framework due to insecure floating-point samplers and disabled local DP in secure aggregation, impacting 87% of macOS Sonoma and 68% of Sequoia data collection.
-
Dithered Gaussian Mechanism for Randomness-Efficient Differential Privacy
The dithered Gaussian mechanism discretizes the output of the Gaussian mechanism via a randomly shifted grid, inheriting Gaussian privacy guarantees while reducing private randomness to a constant per coordinate.
-
Probing Memorization of Tabular In-Context Learning
A new probing framework detects moderate parametric memorization signals in tabular in-context learning models under single-task fine-tuning, strongest on low-cardinality tasks, but signals largely disappear under rea...
-
Auditing Apple's DifferentialPrivacy.framework: Implementation Bugs, Misconfigurations, and Practical Risks
Client-side audit of Apple's closed-source DP framework finds floating-point sampler bugs and misconfigurations that violate DP guarantees in 5 of 9 mechanisms, affecting 87% of data collection on Sonoma and 68% on Sequoia.
-
Assessing Large Language Models for Stabilizing Numerical Expressions in Scientific Software
LLMs match or exceed state-of-the-art traditional methods for stabilizing numerical expressions in scientific software, succeeding on 97.9% of expressions where baselines fail to improve accuracy, but struggle with co...
-
DPSQL+: A Differentially Private SQL Library with a Minimum Frequency Rule
DPSQL+ enforces user-level differential privacy and minimum frequency rules in SQL queries via a modular validator-accountant-backend design, showing practical accuracy on TPC-H workloads and higher query volume under...
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.