PEARL: Data Synthesis via Private Embeddings and Adversarial Reconstruction Learning

TL;DR

PEARL introduces a novel data synthesis framework that uses private embeddings and adversarial learning to generate data with strong privacy guarantees, avoiding the privacy degradation common in iterative methods.

Contribution

The paper presents a one-shot data synthesis method using private embeddings and adversarial reconstruction, providing theoretical guarantees and improved privacy-utility trade-offs over existing approaches.

Findings

Outperforms existing methods at comparable privacy levels

Provides theoretical privacy and performance guarantees

Effective on multiple datasets

Abstract

We propose a new framework of synthesizing data using deep generative models in a differentially private manner. Within our framework, sensitive data are sanitized with rigorous privacy guarantees in a one-shot fashion, such that training deep generative models is possible without re-using the original data. Hence, no extra privacy costs or model constraints are incurred, in contrast to popular approaches such as Differentially Private Stochastic Gradient Descent (DP-SGD), which, among other issues, causes degradation in privacy guarantees as the training iteration increases. We demonstrate a realization of our framework by making use of the characteristic function and an adversarial re-weighting objective, which are of independent interest as well. Our proposal has theoretical guarantees of performance, and empirical evaluations on multiple datasets show that our approach outperforms other methods at reasonable levels of privacy.