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

Constructing adversarial perturbations for deep neural networks is animportant direction of research. Crafting image-dependent adversarialperturbations using white-box feedback has hitherto been the norm for suchadversarial attacks. However, black-box attacks are much more practical forreal-world applications. Universal perturbations applicable across multipleimages are gaining popularity due to their innate generalizability. There havealso been efforts to restrict the perturbations to a few pixels in the image.This helps to retain visual similarity with the original images making suchattacks hard to detect. This paper marks an important step which combines allthese directions of research. We propose the DEceit algorithm for constructingeffective universal pixel-restricted perturbations using only black-boxfeedback from the target network. We conduct empirical investigations using theImageNet validation set on the state-of-the-art deep neural classifiers byvarying the number of pixels to be perturbed from a meagre 10 pixels to as highas all pixels in the image. We find that perturbing only about 10% of thepixels in an image using DEceit achieves a commendable and highly transferableFooling Rate while retaining the visual quality. We further demonstrate thatDEceit can be successfully applied to image dependent attacks as well. In bothsets of experiments, we outperformed several state-of-the-art methods.

A Black-box Adversarial Attack Strategy with Adjustable Sparsity and Generalizability for Deep Image Classifiers