Certified Robustness to Adversarial Examples with Differential Privacy

TL;DR

This paper introduces PixelDP, a scalable certified defense against adversarial examples that leverages differential privacy to provide robustness guarantees for large neural networks across various model types.

Contribution

It presents the first scalable, broad-application certified defense against adversarial attacks using a novel link to differential privacy.

Findings

Successfully scales to large datasets like ImageNet

Supports arbitrary model types

Provides formal robustness guarantees

Abstract

Adversarial examples that fool machine learning models, particularly deep neural networks, have been a topic of intense research interest, with attacks and defenses being developed in a tight back-and-forth. Most past defenses are best effort and have been shown to be vulnerable to sophisticated attacks. Recently a set of certified defenses have been introduced, which provide guarantees of robustness to norm-bounded attacks, but they either do not scale to large datasets or are limited in the types of models they can support. This paper presents the first certified defense that both scales to large networks and datasets (such as Google's Inception network for ImageNet) and applies broadly to arbitrary model types. Our defense, called PixelDP, is based on a novel connection between robustness against adversarial examples and differential privacy, a cryptographically-inspired formalism, that provides a rigorous, generic, and flexible foundation for defense.