Finding Solutions to Generative Adversarial Privacy

TL;DR

This paper introduces heuristics for solving the maximin problem in generative adversarial privacy, including greedy algorithms for linear adversaries and an alternating optimization method for CNN adversaries, to effectively hide private information.

Contribution

It proposes new heuristics for the privacy problem in adversarial settings, improving feature removal and information hiding strategies for linear and CNN adversaries.

Findings

Greedy algorithm improves as data size increases.

Method effectively hides private info while preserving public label accuracy.

Successfully tested on fixed CNN adversaries.

Abstract

We present heuristics for solving the maximin problem induced by the generative adversarial privacy setting for linear and convolutional neural network (CNN) adversaries. In the linear adversary setting, we present a greedy algorithm for approximating the optimal solution for the privatizer, which performs better as the number of instances increases. We also provide an analysis of the algorithm to show that it not only removes the features most correlated with the private label first, but also preserves the prediction accuracy of public labels that are sufficiently independent of the features that are relevant to the private label. In the CNN adversary setting, we present a method of hiding selected information from the adversary while preserving the others through alternately optimizing the goals of the privatizer and the adversary using neural network backpropagation. We experimentally show that our method succeeds on a fixed adversary.