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

Adversarial attack perturbs an image with an imperceptible noise, leading toincorrect model prediction. Recently, a few works showed inherent biasassociated with such attack (robustness bias), where certain subgroups in adataset (e.g. based on class, gender, etc.) are less robust than others. Thisbias not only persists even after adversarial training, but often results insevere performance discrepancies across these subgroups. Existing workscharacterize the subgroup's robustness bias by only checking individualsample's proximity to the decision boundary. In this work, we argue that thismeasure alone is not sufficient and validate our argument via extensiveexperimental analysis. It has been observed that adversarial attacks oftencorrupt the high-frequency components of the input image. We, therefore,propose a holistic approach for quantifying adversarial vulnerability of asample by combining these different perspectives, i.e., degree of model'sreliance on high-frequency features and the (conventional) sample-distance tothe decision boundary. We demonstrate that by reliably estimating adversarialvulnerability at the sample level using the proposed holistic metric, it ispossible to develop a trustworthy system where humans can be alerted about theincoming samples that are highly likely to be misclassified at test time. Thisis achieved with better precision when our holistic metric is used overindividual measures. To further corroborate the utility of the proposedholistic approach, we perform knowledge distillation in a limited-samplesetting. We observe that the student network trained with the subset of samplesselected using our combined metric performs better than both the competingbaselines, viz., where samples are selected randomly or based on theirdistances to the decision boundary.

Holistic Approach to Measure Sample-level Adversarial Vulnerability and its Utility in Building Trustworthy Systems