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

Neural architectures based on attention such as vision transformers arerevolutionizing image recognition. Their main benefit is that attention allowsreasoning about all parts of a scene jointly. In this paper, we show how theglobal reasoning of (scaled) dot-product attention can be the source of a majorvulnerability when confronted with adversarial patch attacks. We provide atheoretical understanding of this vulnerability and relate it to an adversary'sability to misdirect the attention of all queries to a single key token underthe control of the adversarial patch. We propose novel adversarial objectivesfor crafting adversarial patches which target this vulnerability explicitly. Weshow the effectiveness of the proposed patch attacks on popular imageclassification (ViTs and DeiTs) and object detection models (DETR). We findthat adversarial patches occupying 0.5% of the input can lead to robustaccuracies as low as 0% for ViT on ImageNet, and reduce the mAP of DETR on MSCOCO to less than 3%.

Give Me Your Attention: Dot-Product Attention Considered Harmful for Adversarial Patch Robustness