DPSynthesizer: Differentially Private Data Synthesizer for Privacy Preserving Data Sharing.

Area of Science:

• Computer Science
• Data Privacy
• Statistical Analysis

Background:

• Differential privacy is a leading standard for protecting sensitive information in data release.
• Existing methods for generating differentially private synthetic data struggle with high-dimensional and large-domain datasets.
• There is a lack of open-source tools for creating differentially private synthetic data, especially for complex datasets.

Purpose of the Study:

• To introduce DPSynthesizer, a novel open-source toolkit for generating differentially private synthetic data.
• To address the challenges posed by high-dimensional and large-domain data in privacy-preserving data synthesis.
• To provide a practical solution for privacy-preserving data sharing and analytics.

Main Methods:

• The core component, DPCopula, computes differentially private copula functions for high-dimensional data.
• Copula functions model dependencies between variables, enabling the construction of multivariate distributions.
• The toolkit also includes state-of-the-art methods for generating differentially private histograms for lower-dimensional data.

Main Results:

• DPSynthesizer demonstrates feasibility and efficiency in generating differentially private synthetic data.
• The DPCopula method shows promise for handling high-dimensional and large-domain datasets effectively.
• Evaluations across various datasets confirm the utility and performance of the implemented methods.

Conclusions:

• DPSynthesizer provides a valuable open-source solution for privacy-preserving data synthesis.
• The toolkit enhances the ability to share and analyze sensitive data while maintaining strong privacy guarantees.
• It overcomes limitations of previous methods, particularly for complex, high-dimensional datasets.