Defending Against Adversarial Examples by Regularized Deep Embedding

Yao Li, Martin Renqiang Min, Wenchao Yu, Cho-Jui Hsieh, Thomas Lee, Erik Kruus

TL;DR: A general and easy-to-use framework that improves the adversarial robustness of deep classification models through embedding regularization.

Abstract: Recent studies have demonstrated the vulnerability of deep convolutional neural networks against adversarial examples. Inspired by the observation that the intrinsic dimension of image data is much smaller than its pixel space dimension and the vulnerability of neural networks grows with the input dimension, we propose to embed high-dimensional input images into a low-dimensional space to perform classification. However, arbitrarily projecting the input images to a low-dimensional space without regularization will not improve the robustness of deep neural networks. We propose a new framework, Embedding Regularized Classifier (ER-Classifier), which improves the adversarial robustness of the classifier through embedding regularization. Experimental results on several benchmark datasets show that, our proposed framework achieves state-of-the-art performance against strong adversarial attack methods.