Optimal Domain-Aware Privacy Mechanisms for Synthetic Data Generation

2026-07-20

AI Summary

This work addresses the fundamental trade-off between privacy preservation and statistical fidelity in synthetic data generation under differential privacy constraints. It proposes PubMix, the first theoretical framework that directly integrates public data into the core of a differentially private mechanism, rather than merely using it in preprocessing or postprocessing stages. PubMix employs normalized histograms as distribution estimators and leverages domain-specific public data to construct an asymptotically optimal privacy mechanism tailored to the target data domain. By designing a domain-aware noise injection strategy, the method significantly enhances the statistical accuracy and overall utility of the generated synthetic data. Experimental results demonstrate that PubMix consistently outperforms domain-agnostic baselines, achieving markedly improved fidelity while rigorously satisfying differential privacy guarantees.

Abstract

Differential privacy (DP) imposes fundamental trade-offs between privacy and statistical fidelity in synthetic data generation. While access to public data has been shown to improve these trade-offs empirically, existing approaches use public data only indirectly, through pre-processing (e.g., using pre-trained generative models) or post-processing steps (e.g., matching target statistics estimated from public datasets), while relying on domain-agnostic DP mechanisms. In this work, we lay the theoretical framework to study the principled incorporation of public data into DP mechanisms themselves. We consider normalized histograms as distribution estimators and characterize the asymptotically optimal domain-aware privacy mechanism within a specific class of DP mechanisms. We introduce PubMix, a public-data-aware DP mechanism that can be used in histogram-based data synthesis pipelines. Our experiments demonstrate that PubMix significantly improves synthetic data generation quality compared to domain-agnostic privacy mechanisms.

Problem

• differential privacy
• synthetic data generation
• public data
• privacy-utility trade-off
• domain-aware mechanisms

Innovation

• domain-aware privacy
• differential privacy
• synthetic data generation
• public data incorporation
• PubMix