Matches in SemOpenAlex for { <https://semopenalex.org/work/W4313256618> ?p ?o ?g. }
Showing items 1 to 93 of
93
with 100 items per page.
- W4313256618 endingPage "24" @default.
- W4313256618 startingPage "5" @default.
- W4313256618 abstract "Abstract Vision Transformer (ViT) is becoming more and more popular in the field of image processing. This study aims to improve the robustness against the unknown perturbations without retraining the ViT model from scratch. Since our approach does not alter the training phase, it does not need to repeat computationally heavy pretraining of ViT. Specifically, we use test-time adaptation (TTA) for this purpose, which corrects its prediction during test-time by itself. The representative test-time adaptation method, Tent, is recently found to be applicable to ViT by modulating parameters and gradient clipping. However, we observed that Tent sometimes catastrophically fails, especially under severe perturbations. To stabilize the adaptation, we propose a new loss function called Attent, which minimizes the distributional differences of the attention entropy between the source and target. Experiments of image classification task on CIFAR-10-C, CIFAR-100-C, and ImageNet-C show that both Tent and Attent are effective on a wide variety of corruptions. The results also show that by combining Attent and Tent, the classification accuracy on corrupted data is further improved." @default.
- W4313256618 created "2023-01-06" @default.
- W4313256618 creator A5063925941 @default.
- W4313256618 creator A5074059447 @default.
- W4313256618 creator A5089855423 @default.
- W4313256618 date "2022-12-27" @default.
- W4313256618 modified "2023-09-30" @default.
- W4313256618 title "Robustifying Vision Transformer Without Retraining from Scratch Using Attention-Based Test-Time Adaptation" @default.
- W4313256618 cites W1982696459 @default.
- W4313256618 cites W1995875735 @default.
- W4313256618 cites W2108598243 @default.
- W4313256618 cites W2117539524 @default.
- W4313256618 cites W2165698076 @default.
- W4313256618 cites W2194775991 @default.
- W4313256618 cites W2963341956 @default.
- W4313256618 cites W2963351113 @default.
- W4313256618 cites W3035160371 @default.
- W4313256618 cites W3037492894 @default.
- W4313256618 cites W3039883906 @default.
- W4313256618 cites W3095422700 @default.
- W4313256618 cites W3096609285 @default.
- W4313256618 cites W3145185940 @default.
- W4313256618 cites W3177096435 @default.
- W4313256618 cites W4205991051 @default.
- W4313256618 cites W4285602376 @default.
- W4313256618 cites W4312365002 @default.
- W4313256618 doi "https://doi.org/10.1007/s00354-022-00197-9" @default.
- W4313256618 hasPublicationYear "2022" @default.
- W4313256618 type Work @default.
- W4313256618 citedByCount "0" @default.
- W4313256618 crossrefType "journal-article" @default.
- W4313256618 hasAuthorship W4313256618A5063925941 @default.
- W4313256618 hasAuthorship W4313256618A5074059447 @default.
- W4313256618 hasAuthorship W4313256618A5089855423 @default.
- W4313256618 hasBestOaLocation W43132566181 @default.
- W4313256618 hasConcept C104317684 @default.
- W4313256618 hasConcept C111919701 @default.
- W4313256618 hasConcept C119857082 @default.
- W4313256618 hasConcept C120665830 @default.
- W4313256618 hasConcept C121332964 @default.
- W4313256618 hasConcept C139807058 @default.
- W4313256618 hasConcept C144133560 @default.
- W4313256618 hasConcept C153180895 @default.
- W4313256618 hasConcept C154945302 @default.
- W4313256618 hasConcept C155202549 @default.
- W4313256618 hasConcept C165801399 @default.
- W4313256618 hasConcept C185592680 @default.
- W4313256618 hasConcept C2778712577 @default.
- W4313256618 hasConcept C2781235140 @default.
- W4313256618 hasConcept C41008148 @default.
- W4313256618 hasConcept C55493867 @default.
- W4313256618 hasConcept C62520636 @default.
- W4313256618 hasConcept C63479239 @default.
- W4313256618 hasConcept C66322947 @default.
- W4313256618 hasConceptScore W4313256618C104317684 @default.
- W4313256618 hasConceptScore W4313256618C111919701 @default.
- W4313256618 hasConceptScore W4313256618C119857082 @default.
- W4313256618 hasConceptScore W4313256618C120665830 @default.
- W4313256618 hasConceptScore W4313256618C121332964 @default.
- W4313256618 hasConceptScore W4313256618C139807058 @default.
- W4313256618 hasConceptScore W4313256618C144133560 @default.
- W4313256618 hasConceptScore W4313256618C153180895 @default.
- W4313256618 hasConceptScore W4313256618C154945302 @default.
- W4313256618 hasConceptScore W4313256618C155202549 @default.
- W4313256618 hasConceptScore W4313256618C165801399 @default.
- W4313256618 hasConceptScore W4313256618C185592680 @default.
- W4313256618 hasConceptScore W4313256618C2778712577 @default.
- W4313256618 hasConceptScore W4313256618C2781235140 @default.
- W4313256618 hasConceptScore W4313256618C41008148 @default.
- W4313256618 hasConceptScore W4313256618C55493867 @default.
- W4313256618 hasConceptScore W4313256618C62520636 @default.
- W4313256618 hasConceptScore W4313256618C63479239 @default.
- W4313256618 hasConceptScore W4313256618C66322947 @default.
- W4313256618 hasIssue "1" @default.
- W4313256618 hasLocation W43132566181 @default.
- W4313256618 hasOpenAccess W4313256618 @default.
- W4313256618 hasPrimaryLocation W43132566181 @default.
- W4313256618 hasRelatedWork W2065109233 @default.
- W4313256618 hasRelatedWork W2961085424 @default.
- W4313256618 hasRelatedWork W3035617937 @default.
- W4313256618 hasRelatedWork W3172538046 @default.
- W4313256618 hasRelatedWork W4287117966 @default.
- W4313256618 hasRelatedWork W4288725058 @default.
- W4313256618 hasRelatedWork W4306674287 @default.
- W4313256618 hasRelatedWork W4306887858 @default.
- W4313256618 hasRelatedWork W4312845548 @default.
- W4313256618 hasRelatedWork W4385728911 @default.
- W4313256618 hasVolume "41" @default.
- W4313256618 isParatext "false" @default.
- W4313256618 isRetracted "false" @default.
- W4313256618 workType "article" @default.