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

Vision Transformers (ViTs) have a radically different architecture withsignificantly less inductive bias than Convolutional Neural Networks. Alongwith the improvement in performance, security and robustness of ViTs are alsoof great importance to study. In contrast to many recent works that exploit therobustness of ViTs against adversarial examples, this paper investigates arepresentative causative attack, i.e., backdoor. We first examine thevulnerability of ViTs against various backdoor attacks and find that ViTs arealso quite vulnerable to existing attacks. However, we observe that theclean-data accuracy and backdoor attack success rate of ViTs responddistinctively to patch transformations before the positional encoding. Then,based on this finding, we propose an effective method for ViTs to defend bothpatch-based and blending-based trigger backdoor attacks via patch processing.The performances are evaluated on several benchmark datasets, includingCIFAR10, GTSRB, and TinyImageNet, which show the proposed novel defense is verysuccessful in mitigating backdoor attacks for ViTs. To the best of ourknowledge, this paper presents the first defensive strategy that utilizes aunique characteristic of ViTs against backdoor attacks.  The paper will appear in the Proceedings of the AAAI'23 Conference. This workwas initially submitted in November 2021 to CVPR'22, then it was re-submittedto ECCV'22. The paper was made public in June 2022. The authors sincerely thankall the referees from the Program Committees of CVPR'22, ECCV'22, and AAAI'23.

Defending Backdoor Attacks on Vision Transformer via Patch Processing