Detecting Adversarial Data using Perturbation Forgery

TL;DR

This paper introduces Perturbation Forgery, a novel detection method that generalizes well against various unseen adversarial attacks by training on open coverings of noise distributions, addressing limitations of previous techniques.

Contribution

The paper proposes a new adversarial detection approach based on noise distribution perturbation and open covering training, enabling detection of diverse unseen attacks with strong generalization.

Findings

Effective detection of unseen gradient-based, generative, and physical attacks.

Strong generalization demonstrated across multiple datasets and attack types.

Reduced inference overhead compared to existing methods.

Abstract

As a defense strategy against adversarial attacks, adversarial detection aims to identify and filter out adversarial data from the data flow based on discrepancies in distribution and noise patterns between natural and adversarial data. Although previous detection methods achieve high performance in detecting gradient-based adversarial attacks, new attacks based on generative models with imbalanced and anisotropic noise patterns evade detection. Even worse, the significant inference time overhead and limited performance against unseen attacks make existing techniques impractical for real-world use. In this paper, we explore the proximity relationship among adversarial noise distributions and demonstrate the existence of an open covering for these distributions. By training on the open covering of adversarial noise distributions, a detector with strong generalization performance against various types of unseen attacks can be developed. Based on this insight, we heuristically propose Perturbation Forgery, which includes noise distribution perturbation, sparse mask generation, and pseudo-adversarial data production, to train an adversarial detector capable of detecting any unseen gradient-based, generative-based, and physical adversarial attacks. Comprehensive experiments conducted on multiple general and facial datasets, with a wide spectrum of attacks, validate the strong generalization of our method.