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

Deep neural networks have been shown to be vulnerable to adversarial examplesdeliberately constructed to misclassify victim models. As most adversarialexamples have restricted their perturbations to $L_{p}$-norm, existing defensemethods have focused on these types of perturbations and less attention hasbeen paid to unrestricted adversarial examples; which can create more realisticattacks, able to deceive models without affecting human predictions. To addressthis problem, the proposed adversarial attack generates an unrestrictedadversarial example with a limited number of parameters. The attack selectsthree points on the input image and based on their locations transforms theimage into an adversarial example. By limiting the range of movement andlocation of these three points and using a discriminatory network, the proposedunrestricted adversarial example preserves the image appearance. Experimentalresults show that the proposed adversarial examples obtain an average successrate of 93.5% in terms of human evaluation on the MNIST and SVHN datasets. Italso reduces the model accuracy by an average of 73% on six datasets MNIST,FMNIST, SVHN, CIFAR10, CIFAR100, and ImageNet. It should be noted that, in thecase of attacks, lower accuracy in the victim model denotes a more successfulattack. The adversarial train of the attack also improves model robustnessagainst a randomly transformed image.

Generating Unrestricted Adversarial Examples via Three Parameters