Generalization in Generative Adversarial Networks: A Novel Perspective from Privacy Protection

TL;DR

This paper explores the generalization properties of GANs through the lens of privacy protection, demonstrating that differentially private training bounds overfitting and reduces information leakage, with implications for model robustness.

Contribution

It provides a theoretical framework linking privacy protection to GAN generalization, reinterprets existing models, and empirically evaluates information leakage mitigation techniques.

Findings

Differentially private GAN training bounds overfitting.

Lipschitz regularization reduces information leakage.

Reinterpretation of Bayesian GAN based on privacy insights.

Abstract

In this paper, we aim to understand the generalization properties of generative adversarial networks (GANs) from a new perspective of privacy protection. Theoretically, we prove that a differentially private learning algorithm used for training the GAN does not overfit to a certain degree, i.e., the generalization gap can be bounded. Moreover, some recent works, such as the Bayesian GAN, can be re-interpreted based on our theoretical insight from privacy protection. Quantitatively, to evaluate the information leakage of well-trained GAN models, we perform various membership attacks on these models. The results show that previous Lipschitz regularization techniques are effective in not only reducing the generalization gap but also alleviating the information leakage of the training dataset.