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 Tools annual

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 Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Adversarial training (AT) is widely considered as the most promising strategyto defend against adversarial attacks and has drawn increasing interest fromresearchers. However, the existing AT methods still suffer from two challenges.First, they are unable to handle unrestricted adversarial examples (UAEs),which are built from scratch, as opposed to restricted adversarial examples(RAEs), which are created by adding perturbations bound by an $l_p$ norm toobserved examples. Second, the existing AT methods often achieve adversarialrobustness at the expense of standard generalizability (i.e., the accuracy onnatural examples) because they make a tradeoff between them. To overcome thesechallenges, we propose a unique viewpoint that understands UAEs asimperceptibly perturbed unobserved examples. Also, we find that the tradeoffresults from the separation of the distributions of adversarial examples andnatural examples. Based on these ideas, we propose a novel AT approach calledProvable Unrestricted Adversarial Training (PUAT), which can provide a targetclassifier with comprehensive adversarial robustness against both UAE and RAE,and simultaneously improve its standard generalizability. Particularly, PUATutilizes partially labeled data to achieve effective UAE generation byaccurately capturing the natural data distribution through a novel augmentedtriple-GAN. At the same time, PUAT extends the traditional AT by introducingthe supervised loss of the target classifier into the adversarial loss andachieves the alignment between the UAE distribution, the natural datadistribution, and the distribution learned by the classifier, with thecollaboration of the augmented triple-GAN. Finally, the solid theoreticalanalysis and extensive experiments conducted on widely-used benchmarksdemonstrate the superiority of PUAT.

Provable Unrestricted Adversarial Training without Compromise with Generalizability