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 examples are perturbed inputs designed to fool machine learningmodels. Most recent works on adversarial examples for image classificationfocus on directly modifying pixels with minor perturbations. A commonrequirement in all these works is that the malicious perturbations should besmall enough (measured by an L_p norm for some p) so that they areimperceptible to humans. However, small perturbations can be unnecessarilyrestrictive and limit the diversity of adversarial examples generated. Further,an L_p norm based distance metric ignores important structure patterns hiddenin images that are important to human perception. Consequently, even the minorperturbation introduced in recent works often makes the adversarial examplesless natural to humans. More importantly, they often do not transfer well andare therefore less effective when attacking black-box models especially forthose protected by a defense mechanism. In this paper, we propose astructure-preserving transformation (SPT) for generating natural and diverseadversarial examples with extremely high transferability. The key idea of ourapproach is to allow perceptible deviation in adversarial examples whilekeeping structure patterns that are central to a human classifier. Empiricalresults on the MNIST and the fashion-MNIST datasets show that adversarialexamples generated by our approach can easily bypass strong adversarialtraining. Further, they transfer well to other target models with no loss orlittle loss of successful attack rate.

Structure-Preserving Transformation: Generating Diverse and Transferable Adversarial Examples