L-RED: Efficient Post-Training Detection of Imperceptible Backdoor Attacks without Access to the Training Set

TL;DR

This paper introduces L-RED, a computationally efficient method for post-training detection of imperceptible backdoor attacks in neural networks that requires minimal clean data and no knowledge of source classes.

Contribution

L-RED is a novel Lagrangian-based defense that detects backdoor attacks without prior knowledge of source classes or large datasets.

Findings

Detected 56 out of 60 backdoor attacks in experiments.

Requires only two clean images per class for effective detection.

Does not assume all non-target classes are source classes.

Abstract

Backdoor attacks (BAs) are an emerging form of adversarial attack typically against deep neural network image classifiers. The attacker aims to have the classifier learn to classify to a target class when test images from one or more source classes contain a backdoor pattern, while maintaining high accuracy on all clean test images. Reverse-Engineering-based Defenses (REDs) against BAs do not require access to the training set but only to an independent clean dataset. Unfortunately, most existing REDs rely on an unrealistic assumption that all classes except the target class are source classes of the attack. REDs that do not rely on this assumption often require a large set of clean images and heavy computation. In this paper, we propose a Lagrangian-based RED (L-RED) that does not require knowledge of the number of source classes (or whether an attack is present). Our defense requires very few clean images to effectively detect BAs and is computationally efficient. Notably, we detect 56 out of 60 BAs using only two clean images per class in our experiments on CIFAR-10.