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

Black-box adversarial attacks have shown strong potential to subvert machinelearning models. Existing black-box adversarial attacks craft the adversarialexamples by iteratively querying the target model and/or leveraging thetransferability of a local surrogate model. Whether such attack can succeedremains unknown to the adversary when empirically designing the attack. In thispaper, to our best knowledge, we take the first step to study a new paradigm ofadversarial attacks -- certifiable black-box attack that can guarantee theattack success rate of the crafted adversarial examples. Specifically, werevise the randomized smoothing to establish novel theories for ensuring theattack success rate of the adversarial examples. To craft the adversarialexamples with the certifiable attack success rate (CASR) guarantee, we designseveral novel techniques, including a randomized query method to query thetarget model, an initialization method with smoothed self-supervisedperturbation to derive certifiable adversarial examples, and a geometricshifting method to reduce the perturbation size of the certifiable adversarialexamples for better imperceptibility. We have comprehensively evaluated theperformance of the certifiable black-box attack on CIFAR10 and ImageNetdatasets against different levels of defenses. Both theoretical andexperimental results have validated the effectiveness of the proposedcertifiable attack.

Certifiable Black-Box Attack: Ensuring Provably Successful Attack for Adversarial Examples