FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W3176799429> ?p ?o ?g. }

• W3176799429 abstract "Deep visual models are susceptible to adversarial perturbations to inputs. Although these signals are carefully crafted, they still appear noise-like patterns to humans. This observation has led to the argument that deep visual representation is misaligned with human perception. We counter-argue by providing evidence of human-meaningful patterns in adversarial perturbations. We first propose an attack that fools a network to confuse a whole category of objects (source class) with a target label. Our attack also limits the unintended fooling by samples from non-sources classes, thereby circumscribing human-defined semantic notions for network fooling. We show that the proposed attack not only leads to the emergence of regular geometric patterns in the perturbations, but also reveals insightful information about the decision boundaries of deep models. Exploring this phenomenon further, we alter the `adversarial' objective of our attack to use it as a tool to `explain' deep visual representation. We show that by careful channeling and projection of the perturbations computed by our method, we can visualize a model's understanding of human-defined semantic notions. Finally, we exploit the explanability properties of our perturbations to perform image generation, inpainting and interactive image manipulation by attacking adversarialy robust `classifiers'.In all, our major contribution is a novel pragmatic adversarial attack that is subsequently transformed into a tool to interpret the visual models. The article also makes secondary contributions in terms of establishing the utility of our attack beyond the adversarial objective with multiple interesting applications." @default.
• W3176799429 created "2021-07-05" @default.
• W3176799429 creator A5009750573 @default.
• W3176799429 creator A5034919276 @default.
• W3176799429 creator A5054140825 @default.
• W3176799429 creator A5069697936 @default.
• W3176799429 date "2021-06-19" @default.
• W3176799429 modified "2023-09-27" @default.
• W3176799429 title "Attack to Fool and Explain Deep Networks" @default.
• W3176799429 cites W1522301498 @default.
• W3176799429 cites W1686810756 @default.
• W3176799429 cites W1825675169 @default.
• W3176799429 cites W2108598243 @default.
• W3176799429 cites W2133059825 @default.
• W3176799429 cites W2157494358 @default.
• W3176799429 cites W2180612164 @default.
• W3176799429 cites W2183341477 @default.
• W3176799429 cites W2194775991 @default.
• W3176799429 cites W2243397390 @default.
• W3176799429 cites W2295936755 @default.
• W3176799429 cites W2460937040 @default.
• W3176799429 cites W2517229335 @default.
• W3176799429 cites W2543927648 @default.
• W3176799429 cites W2557449848 @default.
• W3176799429 cites W2618098489 @default.
• W3176799429 cites W2736899637 @default.
• W3176799429 cites W2738229973 @default.
• W3176799429 cites W2766108848 @default.
• W3176799429 cites W2768899812 @default.
• W3176799429 cites W2774644650 @default.
• W3176799429 cites W2787970001 @default.
• W3176799429 cites W2797455600 @default.
• W3176799429 cites W2895739182 @default.
• W3176799429 cites W2901107321 @default.
• W3176799429 cites W2913848079 @default.
• W3176799429 cites W2930003153 @default.
• W3176799429 cites W2942630857 @default.
• W3176799429 cites W2945359720 @default.
• W3176799429 cites W2946956149 @default.
• W3176799429 cites W2947874337 @default.
• W3176799429 cites W2948776119 @default.
• W3176799429 cites W2949389295 @default.
• W3176799429 cites W2951954400 @default.
• W3176799429 cites W2952388520 @default.
• W3176799429 cites W2954978443 @default.
• W3176799429 cites W2962700793 @default.
• W3176799429 cites W2962818872 @default.
• W3176799429 cites W2962847335 @default.
• W3176799429 cites W2962872506 @default.
• W3176799429 cites W2962933288 @default.
• W3176799429 cites W2963068442 @default.
• W3176799429 cites W2963148663 @default.
• W3176799429 cites W2963163009 @default.
• W3176799429 cites W2963165363 @default.
• W3176799429 cites W2963178695 @default.
• W3176799429 cites W2963184668 @default.
• W3176799429 cites W2963207607 @default.
• W3176799429 cites W2963561004 @default.
• W3176799429 cites W2963564844 @default.
• W3176799429 cites W2963771536 @default.
• W3176799429 cites W2963809642 @default.
• W3176799429 cites W2963839617 @default.
• W3176799429 cites W2963855547 @default.
• W3176799429 cites W2964116600 @default.
• W3176799429 cites W2964153729 @default.
• W3176799429 cites W2964253222 @default.
• W3176799429 cites W2965470910 @default.
• W3176799429 cites W2968044287 @default.
• W3176799429 cites W2968993450 @default.
• W3176799429 cites W2969542116 @default.
• W3176799429 cites W2970030610 @default.
• W3176799429 cites W2970718407 @default.
• W3176799429 cites W2972986629 @default.
• W3176799429 cites W2979878901 @default.
• W3176799429 cites W2984942115 @default.
• W3176799429 cites W2985390828 @default.
• W3176799429 cites W2988157455 @default.
• W3176799429 cites W2988535323 @default.
• W3176799429 cites W2996507027 @default.
• W3176799429 cites W3020933566 @default.
• W3176799429 cites W3034296472 @default.
• W3176799429 cites W3101609372 @default.
• W3176799429 cites W3115397767 @default.
• W3176799429 doi "https://doi.org/10.48550/arxiv.2106.10606" @default.
• W3176799429 hasPublicationYear "2021" @default.
• W3176799429 type Work @default.
• W3176799429 sameAs 3176799429 @default.
• W3176799429 citedByCount "0" @default.
• W3176799429 crossrefType "posted-content" @default.
• W3176799429 hasAuthorship W3176799429A5009750573 @default.
• W3176799429 hasAuthorship W3176799429A5034919276 @default.
• W3176799429 hasAuthorship W3176799429A5054140825 @default.
• W3176799429 hasAuthorship W3176799429A5069697936 @default.
• W3176799429 hasBestOaLocation W31767994291 @default.
• W3176799429 hasConcept C108583219 @default.
• W3176799429 hasConcept C115961682 @default.
• W3176799429 hasConcept C11727466 @default.
• W3176799429 hasConcept C154945302 @default.
• W3176799429 hasConcept C15744967 @default.
• W3176799429 hasConcept C165696696 @default.

• next