Deep Neural Rejection against Adversarial Examples

TL;DR

This paper introduces a deep neural rejection method to detect adversarial examples by identifying anomalous features at multiple network layers, which is more efficient and effective against adaptive attacks.

Contribution

The proposed approach detects adversarial inputs without needing to generate adversarial examples during training and is less computationally intensive than existing methods.

Findings

Outperforms previous detection methods under adaptive white-box attacks

Does not require adversarial example generation during training

More computationally efficient than competing approaches

Abstract

Despite the impressive performances reported by deep neural networks in different application domains, they remain largely vulnerable to adversarial examples, i.e., input samples that are carefully perturbed to cause misclassification at test time. In this work, we propose a deep neural rejection mechanism to detect adversarial examples, based on the idea of rejecting samples that exhibit anomalous feature representations at different network layers. With respect to competing approaches, our method does not require generating adversarial examples at training time, and it is less computationally demanding. To properly evaluate our method, we define an adaptive white-box attack that is aware of the defense mechanism and aims to bypass it. Under this worst-case setting, we empirically show that our approach outperforms previously-proposed methods that detect adversarial examples by only analyzing the feature representation provided by the output network layer.