Unfolding Local Growth Rate Estimates for (Almost) Perfect Adversarial Detection

TL;DR

This paper introduces a lightweight adversarial detector based on local intrinsic dimensionality (LID) that significantly outperforms existing methods, achieving near-perfect detection rates across multiple datasets and network architectures.

Contribution

It proposes a novel, simple LID-based detection method that reinterprets and adapts existing measures to improve adversarial attack detection performance.

Findings

Surpasses state-of-the-art adversarial detection methods.

Achieves near-perfect F1-scores on several datasets.

Demonstrates robustness across different neural network architectures.

Abstract

Convolutional neural networks (CNN) define the state-of-the-art solution on many perceptual tasks. However, current CNN approaches largely remain vulnerable against adversarial perturbations of the input that have been crafted specifically to fool the system while being quasi-imperceptible to the human eye. In recent years, various approaches have been proposed to defend CNNs against such attacks, for example by model hardening or by adding explicit defence mechanisms. Thereby, a small "detector" is included in the network and trained on the binary classification task of distinguishing genuine data from data containing adversarial perturbations. In this work, we propose a simple and light-weight detector, which leverages recent findings on the relation between networks' local intrinsic dimensionality (LID) and adversarial attacks. Based on a re-interpretation of the LID measure and several simple adaptations, we surpass the state-of-the-art on adversarial detection by a significant margin and reach almost perfect results in terms of F1-score for several networks and datasets. Sources available at: https://github.com/adverML/multiLID