Matches in SemOpenAlex for { <https://semopenalex.org/work/W4320736148> ?p ?o ?g. }
Showing items 1 to 69 of
69
with 100 items per page.
- W4320736148 endingPage "533" @default.
- W4320736148 startingPage "517" @default.
- W4320736148 abstract "Current techniques in deep learning are still unable to train adversarially robust classifiers which perform as well as non-robust ones. In this work, we continue to study the space of loss functions, and show that the choice of loss can affect robustness in highly nonintuitive ways. Specifically, we demonstrate that a surprising choice of loss function can in fact improve adversarial robustness against some attacks. Our loss function encourages accuracy on adversarial examples, and explicitly penalizes accuracy on natural examples. This is inspired by the theoretical and empirical works suggesting a fundamental tradeoff between standard accuracy and adversarial robustness. Our method, NAturally Penalized (NAP) loss, achieves 61.5% robust accuracy on CIFAR-10 with $$varepsilon =8/255$$ perturbations in $$ell _infty $$ (against a PGD-60 adversary with 20 random restarts). This improves over the standard PGD defense by over 3%, against other loss functions proposed in the literature. Although TRADES performs better on CIFAR-10 against Auto-Attack, our approach gets better results on CIFAR-100. Our results thus suggest that significant robustness gains are possible by revisiting training techniques, even without additional data." @default.
- W4320736148 created "2023-02-15" @default.
- W4320736148 creator A5026833693 @default.
- W4320736148 date "2023-01-01" @default.
- W4320736148 modified "2023-09-26" @default.
- W4320736148 title "Improving Adversarial Robustness by Penalizing Natural Accuracy" @default.
- W4320736148 cites W2112796928 @default.
- W4320736148 cites W2194775991 @default.
- W4320736148 cites W2618530766 @default.
- W4320736148 cites W2949736877 @default.
- W4320736148 cites W2962872506 @default.
- W4320736148 cites W2962968216 @default.
- W4320736148 cites W2963518130 @default.
- W4320736148 cites W2963952467 @default.
- W4320736148 cites W2964137095 @default.
- W4320736148 cites W2965130685 @default.
- W4320736148 cites W2992308087 @default.
- W4320736148 cites W3036129194 @default.
- W4320736148 cites W3205297166 @default.
- W4320736148 doi "https://doi.org/10.1007/978-3-031-25056-9_33" @default.
- W4320736148 hasPublicationYear "2023" @default.
- W4320736148 type Work @default.
- W4320736148 citedByCount "0" @default.
- W4320736148 crossrefType "book-chapter" @default.
- W4320736148 hasAuthorship W4320736148A5026833693 @default.
- W4320736148 hasConcept C104317684 @default.
- W4320736148 hasConcept C11413529 @default.
- W4320736148 hasConcept C119857082 @default.
- W4320736148 hasConcept C126255220 @default.
- W4320736148 hasConcept C154945302 @default.
- W4320736148 hasConcept C185592680 @default.
- W4320736148 hasConcept C33923547 @default.
- W4320736148 hasConcept C37736160 @default.
- W4320736148 hasConcept C38652104 @default.
- W4320736148 hasConcept C41008148 @default.
- W4320736148 hasConcept C41065033 @default.
- W4320736148 hasConcept C55493867 @default.
- W4320736148 hasConcept C63479239 @default.
- W4320736148 hasConceptScore W4320736148C104317684 @default.
- W4320736148 hasConceptScore W4320736148C11413529 @default.
- W4320736148 hasConceptScore W4320736148C119857082 @default.
- W4320736148 hasConceptScore W4320736148C126255220 @default.
- W4320736148 hasConceptScore W4320736148C154945302 @default.
- W4320736148 hasConceptScore W4320736148C185592680 @default.
- W4320736148 hasConceptScore W4320736148C33923547 @default.
- W4320736148 hasConceptScore W4320736148C37736160 @default.
- W4320736148 hasConceptScore W4320736148C38652104 @default.
- W4320736148 hasConceptScore W4320736148C41008148 @default.
- W4320736148 hasConceptScore W4320736148C41065033 @default.
- W4320736148 hasConceptScore W4320736148C55493867 @default.
- W4320736148 hasConceptScore W4320736148C63479239 @default.
- W4320736148 hasLocation W43207361481 @default.
- W4320736148 hasOpenAccess W4320736148 @default.
- W4320736148 hasPrimaryLocation W43207361481 @default.
- W4320736148 hasRelatedWork W2230740169 @default.
- W4320736148 hasRelatedWork W2924591601 @default.
- W4320736148 hasRelatedWork W2978686724 @default.
- W4320736148 hasRelatedWork W3193369220 @default.
- W4320736148 hasRelatedWork W4221147656 @default.
- W4320736148 hasRelatedWork W4225161019 @default.
- W4320736148 hasRelatedWork W4290859889 @default.
- W4320736148 hasRelatedWork W4311734044 @default.
- W4320736148 hasRelatedWork W4312326921 @default.
- W4320736148 hasRelatedWork W4317664613 @default.
- W4320736148 isParatext "false" @default.
- W4320736148 isRetracted "false" @default.
- W4320736148 workType "book-chapter" @default.