Matches in SemOpenAlex for { <https://semopenalex.org/work/W2924512726> ?p ?o ?g. }
- W2924512726 abstract "Adversarial training is a training scheme designed to counter adversarial attacks by augmenting the training dataset with adversarial examples. Surprisingly, several studies have observed that loss gradients from adversarially trained DNNs are visually more interpretable than those from standard DNNs. Although this phenomenon is interesting, there are only few works that have offered an explanation. In this paper, we attempted to bridge this gap between adversarial robustness and gradient interpretability. To this end, we identified that loss gradients from adversarially trained DNNs align better with human perception because adversarial training restricts gradients closer to the image manifold. We then demonstrated that adversarial training causes loss gradients to be quantitatively meaningful. Finally, we showed that under the adversarial training framework, there exists an empirical trade-off between test accuracy and loss gradient interpretability and proposed two potential approaches to resolving this trade-off." @default.
- W2924512726 created "2019-04-01" @default.
- W2924512726 creator A5000629292 @default.
- W2924512726 creator A5012574410 @default.
- W2924512726 creator A5059993024 @default.
- W2924512726 date "2019-03-27" @default.
- W2924512726 modified "2023-09-27" @default.
- W2924512726 title "Bridging Adversarial Robustness and Gradient Interpretability" @default.
- W2924512726 cites W1849277567 @default.
- W2924512726 cites W2240067561 @default.
- W2924512726 cites W2513314332 @default.
- W2924512726 cites W2594633041 @default.
- W2924512726 cites W2605409611 @default.
- W2924512726 cites W2618043096 @default.
- W2924512726 cites W2626639386 @default.
- W2924512726 cites W2750384547 @default.
- W2924512726 cites W2785760873 @default.
- W2924512726 cites W2810348861 @default.
- W2924512726 cites W2887603965 @default.
- W2924512726 cites W2891612330 @default.
- W2924512726 cites W2896754385 @default.
- W2924512726 cites W2902031012 @default.
- W2924512726 cites W2913266441 @default.
- W2924512726 cites W2962759300 @default.
- W2924512726 cites W2963207607 @default.
- W2924512726 cites W2963249138 @default.
- W2924512726 cites W2963382180 @default.
- W2924512726 cites W2963510708 @default.
- W2924512726 cites W2963542245 @default.
- W2924512726 cites W2963836885 @default.
- W2924512726 cites W2963857521 @default.
- W2924512726 cites W2964116600 @default.
- W2924512726 cites W2964153729 @default.
- W2924512726 cites W2964167449 @default.
- W2924512726 cites W2964197269 @default.
- W2924512726 cites W3035534932 @default.
- W2924512726 cites W3118608800 @default.
- W2924512726 hasPublicationYear "2019" @default.
- W2924512726 type Work @default.
- W2924512726 sameAs 2924512726 @default.
- W2924512726 citedByCount "13" @default.
- W2924512726 countsByYear W29245127262018 @default.
- W2924512726 countsByYear W29245127262019 @default.
- W2924512726 countsByYear W29245127262020 @default.
- W2924512726 countsByYear W29245127262021 @default.
- W2924512726 countsByYear W29245127262023 @default.
- W2924512726 crossrefType "posted-content" @default.
- W2924512726 hasAuthorship W2924512726A5000629292 @default.
- W2924512726 hasAuthorship W2924512726A5012574410 @default.
- W2924512726 hasAuthorship W2924512726A5059993024 @default.
- W2924512726 hasConcept C104317684 @default.
- W2924512726 hasConcept C119857082 @default.
- W2924512726 hasConcept C154945302 @default.
- W2924512726 hasConcept C174348530 @default.
- W2924512726 hasConcept C185592680 @default.
- W2924512726 hasConcept C2781067378 @default.
- W2924512726 hasConcept C37736160 @default.
- W2924512726 hasConcept C38652104 @default.
- W2924512726 hasConcept C41008148 @default.
- W2924512726 hasConcept C41065033 @default.
- W2924512726 hasConcept C55493867 @default.
- W2924512726 hasConcept C63479239 @default.
- W2924512726 hasConceptScore W2924512726C104317684 @default.
- W2924512726 hasConceptScore W2924512726C119857082 @default.
- W2924512726 hasConceptScore W2924512726C154945302 @default.
- W2924512726 hasConceptScore W2924512726C174348530 @default.
- W2924512726 hasConceptScore W2924512726C185592680 @default.
- W2924512726 hasConceptScore W2924512726C2781067378 @default.
- W2924512726 hasConceptScore W2924512726C37736160 @default.
- W2924512726 hasConceptScore W2924512726C38652104 @default.
- W2924512726 hasConceptScore W2924512726C41008148 @default.
- W2924512726 hasConceptScore W2924512726C41065033 @default.
- W2924512726 hasConceptScore W2924512726C55493867 @default.
- W2924512726 hasConceptScore W2924512726C63479239 @default.
- W2924512726 hasLocation W29245127261 @default.
- W2924512726 hasOpenAccess W2924512726 @default.
- W2924512726 hasPrimaryLocation W29245127261 @default.
- W2924512726 hasRelatedWork W2117539524 @default.
- W2924512726 hasRelatedWork W2194775991 @default.
- W2924512726 hasRelatedWork W2626639386 @default.
- W2924512726 hasRelatedWork W2950095299 @default.
- W2924512726 hasRelatedWork W2962851944 @default.
- W2924512726 hasRelatedWork W2963207607 @default.
- W2924512726 hasRelatedWork W2964116600 @default.
- W2924512726 hasRelatedWork W2964153729 @default.
- W2924512726 hasRelatedWork W2964253222 @default.
- W2924512726 hasRelatedWork W2993049325 @default.
- W2924512726 hasRelatedWork W3012913016 @default.
- W2924512726 hasRelatedWork W3041269312 @default.
- W2924512726 hasRelatedWork W3088536850 @default.
- W2924512726 hasRelatedWork W3132767683 @default.
- W2924512726 hasRelatedWork W3133489578 @default.
- W2924512726 hasRelatedWork W3184728771 @default.
- W2924512726 hasRelatedWork W3194040058 @default.
- W2924512726 hasRelatedWork W3201918604 @default.
- W2924512726 hasRelatedWork W3204747554 @default.
- W2924512726 hasRelatedWork W3208923398 @default.
- W2924512726 isParatext "false" @default.
- W2924512726 isRetracted "false" @default.
- W2924512726 magId "2924512726" @default.