Matches in SemOpenAlex for { <https://semopenalex.org/work/W3092753908> ?p ?o ?g. }
- W3092753908 abstract "Adversarial data augmentation has shown promise for training robust deep neural networks against unforeseen data shifts or corruptions. However, it is difficult to define heuristics to generate effective fictitious target distributions containing hard adversarial perturbations that are largely different from the source distribution. In this paper, we propose a novel and effective regularization term for adversarial data augmentation. We theoretically derive it from the information bottleneck principle, which results in a maximum-entropy formulation. Intuitively, this regularization term encourages perturbing the underlying source distribution to enlarge predictive uncertainty of the current model, so that the generated hard adversarial perturbations can improve the model robustness during training. Experimental results on three standard benchmarks demonstrate that our method consistently outperforms the existing state of the art by a statistically significant margin." @default.
- W3092753908 created "2020-10-22" @default.
- W3092753908 creator A5020733462 @default.
- W3092753908 creator A5022800038 @default.
- W3092753908 creator A5033202592 @default.
- W3092753908 creator A5064437234 @default.
- W3092753908 date "2020-10-15" @default.
- W3092753908 modified "2023-09-23" @default.
- W3092753908 title "Maximum-Entropy Adversarial Data Augmentation for Improved Generalization and Robustness" @default.
- W3092753908 cites W1731081199 @default.
- W3092753908 cites W1959608418 @default.
- W3092753908 cites W1965555277 @default.
- W3092753908 cites W2095705004 @default.
- W3092753908 cites W2099111195 @default.
- W3092753908 cites W2101985079 @default.
- W3092753908 cites W2112796928 @default.
- W3092753908 cites W2114771311 @default.
- W3092753908 cites W2123469175 @default.
- W3092753908 cites W2127090196 @default.
- W3092753908 cites W2135354436 @default.
- W3092753908 cites W2144513243 @default.
- W3092753908 cites W2147800946 @default.
- W3092753908 cites W2148603752 @default.
- W3092753908 cites W2157475639 @default.
- W3092753908 cites W2163605009 @default.
- W3092753908 cites W2164411961 @default.
- W3092753908 cites W2296319761 @default.
- W3092753908 cites W2335728318 @default.
- W3092753908 cites W2511131004 @default.
- W3092753908 cites W2549139847 @default.
- W3092753908 cites W2593768305 @default.
- W3092753908 cites W2600383743 @default.
- W3092753908 cites W2683470288 @default.
- W3092753908 cites W2746314669 @default.
- W3092753908 cites W2763549966 @default.
- W3092753908 cites W2795414739 @default.
- W3092753908 cites W2797548502 @default.
- W3092753908 cites W2803832867 @default.
- W3092753908 cites W2810676804 @default.
- W3092753908 cites W2880214242 @default.
- W3092753908 cites W2889965839 @default.
- W3092753908 cites W2902227449 @default.
- W3092753908 cites W2907520152 @default.
- W3092753908 cites W2943160321 @default.
- W3092753908 cites W2947707615 @default.
- W3092753908 cites W2949736877 @default.
- W3092753908 cites W2962935454 @default.
- W3092753908 cites W2963043696 @default.
- W3092753908 cites W2963060032 @default.
- W3092753908 cites W2963207607 @default.
- W3092753908 cites W2963238274 @default.
- W3092753908 cites W2963263347 @default.
- W3092753908 cites W2963286043 @default.
- W3092753908 cites W2963382180 @default.
- W3092753908 cites W2963389226 @default.
- W3092753908 cites W2963399829 @default.
- W3092753908 cites W2963446712 @default.
- W3092753908 cites W2963539647 @default.
- W3092753908 cites W2963608890 @default.
- W3092753908 cites W2963625477 @default.
- W3092753908 cites W2963685250 @default.
- W3092753908 cites W2963826681 @default.
- W3092753908 cites W2964041897 @default.
- W3092753908 cites W2964059111 @default.
- W3092753908 cites W2964121744 @default.
- W3092753908 cites W2964137095 @default.
- W3092753908 cites W2964160479 @default.
- W3092753908 cites W2964184826 @default.
- W3092753908 cites W2969454448 @default.
- W3092753908 cites W2970018230 @default.
- W3092753908 cites W2970518055 @default.
- W3092753908 cites W2970859221 @default.
- W3092753908 cites W2971028215 @default.
- W3092753908 cites W2979454998 @default.
- W3092753908 cites W2981368934 @default.
- W3092753908 cites W2982402336 @default.
- W3092753908 cites W2992308087 @default.
- W3092753908 cites W2994658979 @default.
- W3092753908 cites W2994931756 @default.
- W3092753908 cites W2995480165 @default.
- W3092753908 cites W2996653965 @default.
- W3092753908 cites W3034203264 @default.
- W3092753908 cites W3034373371 @default.
- W3092753908 cites W3098612954 @default.
- W3092753908 cites W3118608800 @default.
- W3092753908 cites W3128475431 @default.
- W3092753908 cites W601603264 @default.
- W3092753908 hasPublicationYear "2020" @default.
- W3092753908 type Work @default.
- W3092753908 sameAs 3092753908 @default.
- W3092753908 citedByCount "0" @default.
- W3092753908 crossrefType "posted-content" @default.
- W3092753908 hasAuthorship W3092753908A5020733462 @default.
- W3092753908 hasAuthorship W3092753908A5022800038 @default.
- W3092753908 hasAuthorship W3092753908A5033202592 @default.
- W3092753908 hasAuthorship W3092753908A5064437234 @default.
- W3092753908 hasConcept C104317684 @default.
- W3092753908 hasConcept C106301342 @default.
- W3092753908 hasConcept C11413529 @default.
- W3092753908 hasConcept C119857082 @default.