Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Differential Privacy Made Easy

Muhammad Aitsam

TL;DR

This work addresses the practical challenge of applying differential privacy by introducing a simplified DP framework that compares three Python DP libraries and a new beginner-friendly library, GRAM-DP. The approach minimizes user tuning by automating parameter handling and providing straightforward usage for non-experts, while illustrating DP principles through concrete mechanisms and basic queries. Empirical results on the Adult dataset demonstrate how the privacy parameter $\varepsilon$ governs the privacy-utility trade-off and compare library performance, highlighting the practicality of GRAM-DP for accessible DP releases. The paper envisions future expansions to additional mechanisms, unbounded sensitivity handling, more queries, and cross-language support to broaden adoption in industry and research.

Abstract

Data privacy is a major issue for many decades, several techniques have been developed to make sure individuals' privacy but still world has seen privacy failures. In 2006, Cynthia Dwork gave the idea of Differential Privacy which gave strong theoretical guarantees for data privacy. Many companies and research institutes developed differential privacy libraries, but in order to get the differentially private results, users have to tune the privacy parameters. In this paper, we minimized these tune-able parameters. The DP-framework is developed which compares the differentially private results of three Python based DP libraries. We also introduced a new very simple DP library (GRAM-DP), so the people with no background of differential privacy can still secure the privacy of the individuals in the dataset while releasing statistical results in public.

Differential Privacy Made Easy

TL;DR

This work addresses the practical challenge of applying differential privacy by introducing a simplified DP framework that compares three Python DP libraries and a new beginner-friendly library, GRAM-DP. The approach minimizes user tuning by automating parameter handling and providing straightforward usage for non-experts, while illustrating DP principles through concrete mechanisms and basic queries. Empirical results on the Adult dataset demonstrate how the privacy parameter governs the privacy-utility trade-off and compare library performance, highlighting the practicality of GRAM-DP for accessible DP releases. The paper envisions future expansions to additional mechanisms, unbounded sensitivity handling, more queries, and cross-language support to broaden adoption in industry and research.

Abstract

Data privacy is a major issue for many decades, several techniques have been developed to make sure individuals' privacy but still world has seen privacy failures. In 2006, Cynthia Dwork gave the idea of Differential Privacy which gave strong theoretical guarantees for data privacy. Many companies and research institutes developed differential privacy libraries, but in order to get the differentially private results, users have to tune the privacy parameters. In this paper, we minimized these tune-able parameters. The DP-framework is developed which compares the differentially private results of three Python based DP libraries. We also introduced a new very simple DP library (GRAM-DP), so the people with no background of differential privacy can still secure the privacy of the individuals in the dataset while releasing statistical results in public.
Paper Structure (29 sections, 8 equations, 6 figures)

This paper contains 29 sections, 8 equations, 6 figures.

Table of Contents

  1. Introduction
  2. Privacy Failures in Past
  3. AOL Privacy Debacle
  4. Insurance Commission (GIS)
  5. Netflix Prize
  6. Differential Privacy
  7. Definitions of Differential Privacy
  8. $\varepsilon$ - Differential Privacy
  9. ($\varepsilon$, $\delta$) - Differential Privacy
  10. Differential Privacy Mechanisms
  11. Randomized Response Mechanism
  12. Laplace Mechanism
  13. Gaussian Mechanism
  14. Exponential Mechanism
  15. Simplified Differential Privacy (DP) Framework
  16. ...and 14 more sections

Figures (6)

  • Figure 1: Frequency plot of Age and Total working hours from Adults Dataset
  • Figure 2: Count Query results for $hrs-per-week$ column in Adult dataset. The unusual behavior of diffprivlib is due to its mechanism for count query. It consider only non-zero cells in column.
  • Figure 3: Sum Query results for $hrs-per-week$ column in Adult dataset. As we can see the Smartnoise performance is worst as compared to diffprivlib and PyDP.
  • Figure 4: Mean Query results for $age$ column in Adult dataset. Results are as per our expectations for this query. With the increase in $\varepsilon$, error is decreasing for all three libraries.
  • Figure 5: Variance Query results for $age$ column in Adult dataset. Mean Scaled Error, Mean Squared Error and RMSPE shows that error for PyDP is slightly higher.
  • ...and 1 more figures