Matches in SemOpenAlex for { <https://semopenalex.org/work/W3088137196> ?p ?o ?g. }
- W3088137196 abstract "Adversarial training is a defense technique that improves adversarial robustness of a deep neural network (DNN) by including adversarial examples in the training data. In this paper, we identify an overlooked problem of adversarial training in that these adversarial examples often have different semantics than the original data, introducing unintended biases into the model. We hypothesize that such non-semantics-preserving (and resultingly ambiguous) adversarial data harm the robustness of the target models. To mitigate such unintended semantic changes of adversarial examples, we propose semantics-preserving adversarial training (SPAT) which encourages perturbation on the pixels that are shared among all classes when generating adversarial examples in the training stage. Experiment results show that SPAT improves adversarial robustness and achieves state-of-the-art results in CIFAR-10 and CIFAR-100." @default.
- W3088137196 created "2020-10-01" @default.
- W3088137196 creator A5015505425 @default.
- W3088137196 creator A5023255633 @default.
- W3088137196 creator A5053318183 @default.
- W3088137196 date "2020-09-23" @default.
- W3088137196 modified "2023-10-07" @default.
- W3088137196 title "Semantics-Preserving Adversarial Training." @default.
- W3088137196 cites W2119112357 @default.
- W3088137196 cites W2143612262 @default.
- W3088137196 cites W2183341477 @default.
- W3088137196 cites W2194775991 @default.
- W3088137196 cites W2257979135 @default.
- W3088137196 cites W2401231614 @default.
- W3088137196 cites W2786104118 @default.
- W3088137196 cites W2911634294 @default.
- W3088137196 cites W2913266441 @default.
- W3088137196 cites W2945033152 @default.
- W3088137196 cites W2945359720 @default.
- W3088137196 cites W2963143631 @default.
- W3088137196 cites W2963207607 @default.
- W3088137196 cites W2963249138 @default.
- W3088137196 cites W2963341956 @default.
- W3088137196 cites W2963496101 @default.
- W3088137196 cites W2963564844 @default.
- W3088137196 cites W2963857521 @default.
- W3088137196 cites W2964116600 @default.
- W3088137196 cites W2964153729 @default.
- W3088137196 cites W2964159205 @default.
- W3088137196 cites W2964253222 @default.
- W3088137196 cites W2970088379 @default.
- W3088137196 cites W2970680991 @default.
- W3088137196 cites W2996308572 @default.
- W3088137196 cites W2996344901 @default.
- W3088137196 cites W2996568780 @default.
- W3088137196 cites W3034672496 @default.
- W3088137196 cites W3034758058 @default.
- W3088137196 cites W3037742886 @default.
- W3088137196 cites W3103340107 @default.
- W3088137196 cites W3118608800 @default.
- W3088137196 hasPublicationYear "2020" @default.
- W3088137196 type Work @default.
- W3088137196 sameAs 3088137196 @default.
- W3088137196 citedByCount "0" @default.
- W3088137196 crossrefType "posted-content" @default.
- W3088137196 hasAuthorship W3088137196A5015505425 @default.
- W3088137196 hasAuthorship W3088137196A5023255633 @default.
- W3088137196 hasAuthorship W3088137196A5053318183 @default.
- W3088137196 hasConcept C104317684 @default.
- W3088137196 hasConcept C108583219 @default.
- W3088137196 hasConcept C119857082 @default.
- W3088137196 hasConcept C154945302 @default.
- W3088137196 hasConcept C15744967 @default.
- W3088137196 hasConcept C184337299 @default.
- W3088137196 hasConcept C185592680 @default.
- W3088137196 hasConcept C199360897 @default.
- W3088137196 hasConcept C204321447 @default.
- W3088137196 hasConcept C2777363581 @default.
- W3088137196 hasConcept C2984842247 @default.
- W3088137196 hasConcept C37736160 @default.
- W3088137196 hasConcept C41008148 @default.
- W3088137196 hasConcept C50644808 @default.
- W3088137196 hasConcept C51632099 @default.
- W3088137196 hasConcept C55493867 @default.
- W3088137196 hasConcept C63479239 @default.
- W3088137196 hasConcept C77805123 @default.
- W3088137196 hasConceptScore W3088137196C104317684 @default.
- W3088137196 hasConceptScore W3088137196C108583219 @default.
- W3088137196 hasConceptScore W3088137196C119857082 @default.
- W3088137196 hasConceptScore W3088137196C154945302 @default.
- W3088137196 hasConceptScore W3088137196C15744967 @default.
- W3088137196 hasConceptScore W3088137196C184337299 @default.
- W3088137196 hasConceptScore W3088137196C185592680 @default.
- W3088137196 hasConceptScore W3088137196C199360897 @default.
- W3088137196 hasConceptScore W3088137196C204321447 @default.
- W3088137196 hasConceptScore W3088137196C2777363581 @default.
- W3088137196 hasConceptScore W3088137196C2984842247 @default.
- W3088137196 hasConceptScore W3088137196C37736160 @default.
- W3088137196 hasConceptScore W3088137196C41008148 @default.
- W3088137196 hasConceptScore W3088137196C50644808 @default.
- W3088137196 hasConceptScore W3088137196C51632099 @default.
- W3088137196 hasConceptScore W3088137196C55493867 @default.
- W3088137196 hasConceptScore W3088137196C63479239 @default.
- W3088137196 hasConceptScore W3088137196C77805123 @default.
- W3088137196 hasLocation W30881371961 @default.
- W3088137196 hasOpenAccess W3088137196 @default.
- W3088137196 hasPrimaryLocation W30881371961 @default.
- W3088137196 hasRelatedWork W2802379474 @default.
- W3088137196 hasRelatedWork W2892128688 @default.
- W3088137196 hasRelatedWork W2894012075 @default.
- W3088137196 hasRelatedWork W2924566522 @default.
- W3088137196 hasRelatedWork W2949384876 @default.
- W3088137196 hasRelatedWork W2951526967 @default.
- W3088137196 hasRelatedWork W2962550381 @default.
- W3088137196 hasRelatedWork W2969312451 @default.
- W3088137196 hasRelatedWork W2972816625 @default.
- W3088137196 hasRelatedWork W2996568780 @default.
- W3088137196 hasRelatedWork W2998135039 @default.
- W3088137196 hasRelatedWork W3022223508 @default.
- W3088137196 hasRelatedWork W3022473380 @default.