Class-Disentanglement and Applications in Adversarial Detection and Defense

What is the minimum necessary information required by a neural net D(・) from an image x to accurately predict its class? Extracting such information in the input space from x can allocate the areas D(・) mainly attending to and shed novel insights to the detection and defense of adversarial attacks. In this paper, we propose "class-disentanglement" that trains a variational autoencoder G(・) to extract this class-dependent information as x − G(x) via a trade-off between reconstructing x by G(x) and classifying x by D(x − G(x)), where the former competes with the latter in decomposing x so the latter retains only necessary information for classification in x − G(x). We apply it to both clean images and their adversarial images and discover that the perturbations generated by adversarial attacks mainly lie in the class-dependent part x − G(x). The decomposition results also provide novel interpretations to classification and attack models. Inspired by these observations, we propose to conduct adversarial detection and adversarial defense respectively on x − G(x) and G(x), which consistently outperform the results on the original x. In experiments, this simple approach substantially improves the detection and defense against different types of adversarial attacks. Code is available: https://github.com/kai-wen-yang/CD-VAE.