Processing math: 100%
Skip to yearly menu bar Skip to main content


Poster

A Unified, Resilient, and Explainable Adversarial Patch Detector

Vishesh Kumar · Akshay Agarwal

ExHall D Poster #406
[ ] [ Project Page ]
Sun 15 Jun 2 p.m. PDT — 4 p.m. PDT

Abstract:

Deep Neural Networks (DNNs), backbone architecture in almosteverycomputervisiontask,arev̲rabadversarialaacks,partica̲rlyphysicalout-of-distribution(OOD)adversarialpatches.Eξstgoftenstruggwitherpretgtheseaackswaystˆalignwithhumanvisualperception.OurosedAdvPatchXAIroducesaralized,robust,andexplaabdefensealgorithmspecificallydesigddefendDsagastphysicaladversarialthreats.AdvPatchXAIemploysanovelpatchdecorrelationlosstˆreducesfeatureredundancyandenhancesthedistctivessofpatchrepresentations,enablgbeerralizationacrossunseenadversarialscenarios.Itarnsprotyπcalpartsaself-rvisedfashion,enhancgerprηbilityandcorrelationwithhumanvision.Themodelutilizesasparselearlayerforclassification,makgthedecision-makgprocessgloballyerprηbthroughasetofardprotypesandlocallyexplaabbyπnpgrevantprotypeswithanima.OurcomprehensiveevaluationshowstˆAdvPatchXAI¬onlyclosesthe`semantic'' gap between latent space and pixel space but also effectively handles unseen adversarial patches even perturbed with unseen corruptions, thereby significantly advancing DNN robustness in practical settings.

Live content is unavailable. Log in and register to view live content