Adversarial Detector with Robust Classifier

TL;DR

This paper introduces a new adversarial detection method using a combination of robust and plain classifiers, which effectively identifies adversarial examples by analyzing their logits, outperforming existing detectors.

Contribution

A novel adversarial detector leveraging both robust and plain classifiers' logits, improving detection accuracy without requiring a fully robust classifier.

Findings

Outperforms state-of-the-art detectors in experiments

Effective detection of adversarial examples using logits analysis

Does not require a fully robust classifier

Abstract

Deep neural network (DNN) models are wellknown to easily misclassify prediction results by using input images with small perturbations, called adversarial examples. In this paper, we propose a novel adversarial detector, which consists of a robust classifier and a plain one, to highly detect adversarial examples. The proposed adversarial detector is carried out in accordance with the logits of plain and robust classifiers. In an experiment, the proposed detector is demonstrated to outperform a state-of-the-art detector without any robust classifier.