English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

In this paper, we study fast training of adversarially robust models. Fromthe analyses of the state-of-the-art defense method, i.e., the multi-stepadversarial training, we hypothesize that the gradient magnitude links to themodel robustness. Motivated by this, we propose to perturb both the image andthe label during training, which we call Bilateral Adversarial Training (BAT).To generate the adversarial label, we derive an closed-form heuristic solution.To generate the adversarial image, we use one-step targeted attack with thetarget label being the most confusing class. In the experiment, we first showthat random start and the most confusing target attack effectively prevent thelabel leaking and gradient masking problem. Then coupled with the adversariallabel part, our model significantly improves the state-of-the-art results. Forexample, against PGD100 white-box attack with cross-entropy loss, on CIFAR10,we achieve 63.7\% versus 47.2\%; on SVHN, we achieve 59.1\% versus 42.1\%. Atlast, the experiment on the very (computationally) challenging ImageNet datasetfurther demonstrates the effectiveness of our fast method.

Bilateral Adversarial Training: Towards Fast Training of More Robust Models Against Adversarial Attacks