Matches in SemOpenAlex for { <https://semopenalex.org/work/W3089774673> ?p ?o ?g. }
- W3089774673 endingPage "6230" @default.
- W3089774673 startingPage "6222" @default.
- W3089774673 abstract "Adversarial training is so far the most effective strategy in defending against adversarial examples. However, it suffers from high computational costs due to the iterative adversarial attacks in each training step. Recent studies show that it is possible to achieve fast Adversarial Training by performing a single-step attack with random initialization. However, such an approach still lags behind state-of-the-art adversarial training algorithms on both stability and model robustness. In this work, we develop a new understanding towards Fast Adversarial Training, by viewing random initialization as performing randomized smoothing for better optimization of the inner maximization problem. Following this new perspective, we also propose a new initialization strategy, backward smoothing, to further improve the stability and model robustness over single-step robust training methods. Experiments on multiple benchmarks demonstrate that our method achieves similar model robustness as the original TRADES method while using much less training time (~3x improvement with the same training schedule)." @default.
- W3089774673 created "2020-10-08" @default.
- W3089774673 creator A5006335513 @default.
- W3089774673 creator A5026746295 @default.
- W3089774673 creator A5051448391 @default.
- W3089774673 creator A5066666034 @default.
- W3089774673 creator A5077322975 @default.
- W3089774673 date "2022-06-28" @default.
- W3089774673 modified "2023-09-30" @default.
- W3089774673 title "Efficient Robust Training via Backward Smoothing" @default.
- W3089774673 cites W2108598243 @default.
- W3089774673 cites W2180612164 @default.
- W3089774673 cites W2194775991 @default.
- W3089774673 cites W2243397390 @default.
- W3089774673 cites W2401231614 @default.
- W3089774673 cites W2460937040 @default.
- W3089774673 cites W2746600820 @default.
- W3089774673 cites W2763421725 @default.
- W3089774673 cites W2765233338 @default.
- W3089774673 cites W2765384636 @default.
- W3089774673 cites W2787496614 @default.
- W3089774673 cites W2787708942 @default.
- W3089774673 cites W2787733970 @default.
- W3089774673 cites W2798801120 @default.
- W3089774673 cites W2911634294 @default.
- W3089774673 cites W2912237282 @default.
- W3089774673 cites W2945793108 @default.
- W3089774673 cites W2946814535 @default.
- W3089774673 cites W2962720772 @default.
- W3089774673 cites W2962729158 @default.
- W3089774673 cites W2963070423 @default.
- W3089774673 cites W2963158386 @default.
- W3089774673 cites W2963207607 @default.
- W3089774673 cites W2963857521 @default.
- W3089774673 cites W2963920068 @default.
- W3089774673 cites W2964054038 @default.
- W3089774673 cites W2964153729 @default.
- W3089774673 cites W2964253222 @default.
- W3089774673 cites W2970049488 @default.
- W3089774673 cites W2970317235 @default.
- W3089774673 cites W2970680991 @default.
- W3089774673 cites W2971109239 @default.
- W3089774673 cites W2971316968 @default.
- W3089774673 cites W2991240978 @default.
- W3089774673 cites W2994702848 @default.
- W3089774673 cites W2996344901 @default.
- W3089774673 cites W2996564870 @default.
- W3089774673 cites W2996950444 @default.
- W3089774673 cites W3008105014 @default.
- W3089774673 cites W3033210711 @default.
- W3089774673 cites W3034994123 @default.
- W3089774673 cites W3035345420 @default.
- W3089774673 cites W3080297477 @default.
- W3089774673 cites W3098632754 @default.
- W3089774673 cites W3118608800 @default.
- W3089774673 doi "https://doi.org/10.1609/aaai.v36i6.20571" @default.
- W3089774673 hasPublicationYear "2022" @default.
- W3089774673 type Work @default.
- W3089774673 sameAs 3089774673 @default.
- W3089774673 citedByCount "10" @default.
- W3089774673 countsByYear W30897746732020 @default.
- W3089774673 countsByYear W30897746732021 @default.
- W3089774673 countsByYear W30897746732023 @default.
- W3089774673 crossrefType "journal-article" @default.
- W3089774673 hasAuthorship W3089774673A5006335513 @default.
- W3089774673 hasAuthorship W3089774673A5026746295 @default.
- W3089774673 hasAuthorship W3089774673A5051448391 @default.
- W3089774673 hasAuthorship W3089774673A5066666034 @default.
- W3089774673 hasAuthorship W3089774673A5077322975 @default.
- W3089774673 hasBestOaLocation W30897746731 @default.
- W3089774673 hasConcept C104317684 @default.
- W3089774673 hasConcept C111919701 @default.
- W3089774673 hasConcept C114466953 @default.
- W3089774673 hasConcept C119857082 @default.
- W3089774673 hasConcept C121332964 @default.
- W3089774673 hasConcept C126255220 @default.
- W3089774673 hasConcept C153294291 @default.
- W3089774673 hasConcept C154945302 @default.
- W3089774673 hasConcept C185592680 @default.
- W3089774673 hasConcept C199360897 @default.
- W3089774673 hasConcept C2776330181 @default.
- W3089774673 hasConcept C2777211547 @default.
- W3089774673 hasConcept C31972630 @default.
- W3089774673 hasConcept C33923547 @default.
- W3089774673 hasConcept C3770464 @default.
- W3089774673 hasConcept C37736160 @default.
- W3089774673 hasConcept C41008148 @default.
- W3089774673 hasConcept C55493867 @default.
- W3089774673 hasConcept C63479239 @default.
- W3089774673 hasConcept C68387754 @default.
- W3089774673 hasConceptScore W3089774673C104317684 @default.
- W3089774673 hasConceptScore W3089774673C111919701 @default.
- W3089774673 hasConceptScore W3089774673C114466953 @default.
- W3089774673 hasConceptScore W3089774673C119857082 @default.
- W3089774673 hasConceptScore W3089774673C121332964 @default.
- W3089774673 hasConceptScore W3089774673C126255220 @default.
- W3089774673 hasConceptScore W3089774673C153294291 @default.
- W3089774673 hasConceptScore W3089774673C154945302 @default.