Generative Adversarial Privacy

TL;DR

This paper introduces generative adversarial privacy (GAP), a data-driven framework inspired by GANs that learns privacy mechanisms through a minimax game, providing privacy guarantees and demonstrating effectiveness on face data.

Contribution

GAP is a novel framework that formulates privacy mechanism design as a minimax game, enabling data-driven learning with theoretical privacy guarantees.

Findings

GAP effectively learns privatization mechanisms from data.

GAP provides privacy guarantees against strong adversaries.

Performance evaluated on GENKI face database shows promising results.

Abstract

We present a data-driven framework called generative adversarial privacy (GAP). Inspired by recent advancements in generative adversarial networks (GANs), GAP allows the data holder to learn the privatization mechanism directly from the data. Under GAP, finding the optimal privacy mechanism is formulated as a constrained minimax game between a privatizer and an adversary. We show that for appropriately chosen adversarial loss functions, GAP provides privacy guarantees against strong information-theoretic adversaries. We also evaluate GAP's performance on the GENKI face database.