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

While the transferability property of adversarial examples allows theadversary to perform black-box attacks (i.e., the attacker has no knowledgeabout the target model), the transfer-based adversarial attacks have gainedgreat attention. Previous works mostly study gradient variation or imagetransformations to amplify the distortion on critical parts of inputs. Thesemethods can work on transferring across models with limited differences, i.e.,from CNNs to CNNs, but always fail in transferring across models with widedifferences, such as from CNNs to ViTs. Alternatively, model ensembleadversarial attacks are proposed to fuse outputs from surrogate models withdiverse architectures to get an ensemble loss, making the generated adversarialexample more likely to transfer to other models as it can fool multiple modelsconcurrently. However, existing ensemble attacks simply fuse the outputs of thesurrogate models evenly, thus are not efficacious to capture and amplify theintrinsic transfer information of adversarial examples. In this paper, wepropose an adaptive ensemble attack, dubbed AdaEA, to adaptively control thefusion of the outputs from each model, via monitoring the discrepancy ratio oftheir contributions towards the adversarial objective. Furthermore, an extradisparity-reduced filter is introduced to further synchronize the updatedirection. As a result, we achieve considerable improvement over the existingensemble attacks on various datasets, and the proposed AdaEA can also boostexisting transfer-based attacks, which further demonstrates its efficacy andversatility.

An Adaptive Model Ensemble Adversarial Attack for Boosting Adversarial Transferability