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

In generating adversarial examples, the conventional black-box attack methodsrely on sufficient feedback from the to-be-attacked models by repeatedlyquerying until the attack is successful, which usually results in thousands oftrials during an attack. This may be unacceptable in real applications sinceMachine Learning as a Service Platform (MLaaS) usually only returns the finalresult (i.e., hard-label) to the client and a system equipped with certaindefense mechanisms could easily detect malicious queries. By contrast, afeasible way is a hard-label attack that simulates an attacked action beingpermitted to conduct a limited number of queries. To implement this idea, inthis paper, we bypass the dependency on the to-be-attacked model and benefitfrom the characteristics of the distributions of adversarial examples toreformulate the attack problem in a distribution transform manner and propose adistribution transform-based attack (DTA). DTA builds a statistical mappingfrom the benign example to its adversarial counterparts by tackling theconditional likelihood under the hard-label black-box settings. In this way, itis no longer necessary to query the target model frequently. A well-trained DTAmodel can directly and efficiently generate a batch of adversarial examples fora certain input, which can be used to attack un-seen models based on theassumed transferability. Furthermore, we surprisingly find that thewell-trained DTA model is not sensitive to the semantic spaces of the trainingdataset, meaning that the model yields acceptable attack performance on otherdatasets. Extensive experiments validate the effectiveness of the proposed ideaand the superiority of DTA over the state-of-the-art.

DTA: Distribution Transform-based Attack for Query-Limited Scenario