Differential Privacy (DP) is a cornerstone for ensuring privacy in data analysis by injecting carefully calibrated noise into statistical queries. While numerous DP tools focus on privacy protection, few provide accuracy information, specially for data-dependent computations like averages or quotients of DP-sums. This paper introduces a novel approach to compute confidence intervals, i.e., α-β accuracy, for these computations, leveraging principles from uncertainty propagation. Our method identifies conditions under which analytical error can be predicted, revealing two key invariants: the analytical error improves with large dataset sizes, and addition of values with higher variability require larger dataset sizes for accurate estimation. To simplify adoption, we also propose accuracy tuners to enable rapid determination of minimum dataset sizes and explore trade-offs between privacy budgets and the possibility to perform accuracy estimations. Our theoretical contributions are validated through an empirical evaluation that explores the applicability of fractional uncertainties for computing concrete α-β error across diverse scenarios.

The work of Fuzz has pioneered the use of functional programming languages where types allow reasoning about the sensitivity of programs. Fuzz and subsequent work (e.g., DFuzz and Duet) use advanced technical devices like linear types, modal types, and partial evaluation. These features usually require the design of a new programming language from scratch—a significant task on its own! While these features are part of the classical toolbox of programming languages, they are often unfamiliar to non-experts in this field. Fortunately, recent studies (e.g., Solo ) have shown that linear and complex types in general, are not strictly needed for the task of determining programs’ sensitivity since this can be achieved by annotating base types with static sensitivity information. In this work, we take a different approach. We propose to enrich base types with information about the metric relation between values, and we present the novel idea of applying parametricity to derive direct proofs for the sensitivity of functions. A direct consequence of our result is that calculating and proving the sensitivity of functions is reduced to simply type-checking in a programming language with support for polymorphism and type-level naturals. We formalize our main result in a calculus, prove its soundness, and implement a software library in the programming language Haskell-where we reason about the sensitivity of canonical examples. We show that the simplicity of our approach allows us to exploit the type inference of the host language to support a limited form of sensitivity inference. Furthermore, we extend the language with a privacy monad to showcase how our library can be used in practical scenarios such as the implementation of differentially private programs, where the privacy guarantees depend on the sensitivity of user-defined functions. Our library, called Spar , is implemented in less than 500 lines of code.

Differential privacy offers a formal framework for reasoning about the privacy and accuracy of computations on private data. It also offers a rich set of building blocks for constructing private data analyses. When carefully calibrated, these analyses simultaneously guarantee the privacy of the individuals contributing their data, and the accuracy of the data analysis results, inferring useful properties about the population. The compositional nature of differential privacy has motivated the design and implementation of several programming languages to ease the implementation of differentially private analyses. Even though these programming languages provide support for reasoning about privacy, most of them disregard reasoning about the accuracy of data analyses. To overcome this limitation, we present DPella, a programming framework providing data analysts with support for reasoning about privacy, accuracy, and their trade-offs. The distinguishing feature of DPella is a novel component that statically tracks the accuracy of different data analyses. To provide tight accuracy estimations, this component leverages taint analysis for automatically inferring statistical independence of the different noise quantities added for guaranteeing privacy. We evaluate our approach by implementing several classical queries from the literature and showing how data analysts can calibrate the privacy parameters to meet the accuracy requirements, and vice versa.

Differential privacy offers a formal framework for reasoning about privacy and accuracy of computations on private data. It also offers a rich set of building blocks for constructing private data analyses. When carefully calibrated, these analyses simultaneously guarantee the privacy of the individuals contributing their data, and the accuracy of the data analyses results, inferring useful properties about the population. The compositional nature of differential privacy has motivated the design and implementation of several programming languages aimed at helping a data analyst in programming differentially private analyses. However, most of the programming languages for differential privacy proposed so far provide support for reasoning about privacy but not for reasoning about the accuracy of data analyses. To overcome this limitation, in this work we present DPella, a programming framework providing data analysts with support for reasoning about privacy, accuracy and their trade-offs. The distinguishing feature of DPella is a novel component which statically tracks the accuracy of different data analyses. In order to make tighter accuracy estimations, this component leverages taint analysis for automatically inferring statistical independence of the different noise quantities added for guaranteeing privacy. We evaluate our approach by implementing several classical queries from the literature and showing how data analysts can figure out the best manner to calibrate privacy to meet the accuracy requirements.