Differentially Private Synthesization of Multi-Dimensional Data using Copula Functions.

Area of Science:

• Computer Science
• Data Privacy
• Statistical Modeling

Background:

• Differential privacy is crucial for secure statistical data release.
• Existing methods struggle with high-dimensional data due to error and complexity.
• Need for privacy-preserving techniques for complex, multi-dimensional datasets.

Purpose of the Study:

• To introduce DPCopula, a novel differentially private data synthesization technique.
• To address limitations of current methods for high-dimensional data.
• To generate accurate synthetic multi-dimensional data with strong privacy.

Main Methods:

• Utilizing Copula functions to model dependencies in multivariate data.
• Implementing differentially private estimation of copula parameters (MLE and Kendall's τ).
• Sampling synthetic data from the estimated private copula function.

Main Results:

• Formal proofs of privacy guarantees and convergence properties.
• Demonstrated high accuracy in synthetic multi-dimensional data generation.
• Outperformed state-of-the-art techniques in data utility.

Conclusions:

• DPCopula effectively generates high-utility synthetic multi-dimensional data under differential privacy.
• The method overcomes limitations of existing techniques for complex datasets.
• Offers a robust solution for private statistical data release.