FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W4385801322> ?p ?o ?g. }

• W4385801322 abstract "Data augmentation is a promising way to enhance the generalization ability of deep learning models. Many proxy-free and proxy-based automated augmentation methods are proposed to search for the best augmentation for target datasets. However, the proxy-free methods require lots of searching overhead, while the proxy-based methods introduce optimization gaps with the actual task. In this paper, we explore a new proxy-free approach that only needs a small number of searches (~ 5 vs 100 of RandAugment) to alleviate these issues. Specifically, we propose Adaptive Automated Augmentation (A <sup xmlns:mml=http://www.w3.org/1998/Math/MathML xmlns:xlink=http://www.w3.org/1999/xlink>2</sup> -Aug), a simple and effective proxy-free framework, which seeks to mine the adaptive ensemble knowledge of multiple augmentations to further improve the adaptability of each candidate augmentation. Firstly, A <sup xmlns:mml=http://www.w3.org/1998/Math/MathML xmlns:xlink=http://www.w3.org/1999/xlink>2</sup> -Aug automatically learns the ensemble logit from multiple candidate augmentations, which is jointly optimized and adaptive to target tasks. Secondly, the adaptive ensemble logit is used to distill each logit of input augmentation via KL divergence. In this way, these a few candidate augmentations can implicitly learn strong adaptability for the target datasets, which enjoy similar effects with many searches of RandAugment. Finally, equipped with joint training via separate BatchNorm and normalized distillation, A <sup xmlns:mml=http://www.w3.org/1998/Math/MathML xmlns:xlink=http://www.w3.org/1999/xlink>2</sup> -Aug obtains state-of-the-art performance with less training budget. In experiments, our A <sup xmlns:mml=http://www.w3.org/1998/Math/MathML xmlns:xlink=http://www.w3.org/1999/xlink>2</sup> -Aug achieves 4% performance gain on CIFAR-100, which substantially outperforms other methods. On ImageNet, we obtain a top-1 accuracy of 79.2% for ResNet-50, a 1.6% boosting over the AutoAugment with at least 25× faster training speed." @default.
• W4385801322 created "2023-08-15" @default.
• W4385801322 creator A5040367049 @default.
• W4385801322 creator A5085482371 @default.
• W4385801322 date "2023-06-01" @default.
• W4385801322 modified "2023-09-27" @default.
• W4385801322 title "A²-Aug: Adaptive Automated Data Augmentation" @default.
• W4385801322 cites W1903029394 @default.
• W4385801322 cites W2102605133 @default.
• W4385801322 cites W2108598243 @default.
• W4385801322 cites W2194775991 @default.
• W4385801322 cites W2739879705 @default.
• W4385801322 cites W2900595477 @default.
• W4385801322 cites W2949736877 @default.
• W4385801322 cites W2962710014 @default.
• W4385801322 cites W2963140444 @default.
• W4385801322 cites W2964081807 @default.
• W4385801322 cites W2987385519 @default.
• W4385801322 cites W2992308087 @default.
• W4385801322 cites W3009184501 @default.
• W4385801322 cites W3035682985 @default.
• W4385801322 cites W3108124733 @default.
• W4385801322 cites W3108411658 @default.
• W4385801322 cites W3108491103 @default.
• W4385801322 cites W3108970435 @default.
• W4385801322 cites W3162847008 @default.
• W4385801322 doi "https://doi.org/10.1109/cvprw59228.2023.00221" @default.
• W4385801322 hasPublicationYear "2023" @default.
• W4385801322 type Work @default.
• W4385801322 citedByCount "0" @default.
• W4385801322 crossrefType "proceedings-article" @default.
• W4385801322 hasAuthorship W4385801322A5040367049 @default.
• W4385801322 hasAuthorship W4385801322A5085482371 @default.
• W4385801322 hasConcept C119857082 @default.
• W4385801322 hasConcept C124101348 @default.
• W4385801322 hasConcept C134306372 @default.
• W4385801322 hasConcept C142724271 @default.
• W4385801322 hasConcept C154945302 @default.
• W4385801322 hasConcept C177148314 @default.
• W4385801322 hasConcept C177606310 @default.
• W4385801322 hasConcept C18903297 @default.
• W4385801322 hasConcept C2780148112 @default.
• W4385801322 hasConcept C2982777018 @default.
• W4385801322 hasConcept C33923547 @default.
• W4385801322 hasConcept C41008148 @default.
• W4385801322 hasConcept C535046627 @default.
• W4385801322 hasConcept C71924100 @default.
• W4385801322 hasConcept C86803240 @default.
• W4385801322 hasConceptScore W4385801322C119857082 @default.
• W4385801322 hasConceptScore W4385801322C124101348 @default.
• W4385801322 hasConceptScore W4385801322C134306372 @default.
• W4385801322 hasConceptScore W4385801322C142724271 @default.
• W4385801322 hasConceptScore W4385801322C154945302 @default.
• W4385801322 hasConceptScore W4385801322C177148314 @default.
• W4385801322 hasConceptScore W4385801322C177606310 @default.
• W4385801322 hasConceptScore W4385801322C18903297 @default.
• W4385801322 hasConceptScore W4385801322C2780148112 @default.
• W4385801322 hasConceptScore W4385801322C2982777018 @default.
• W4385801322 hasConceptScore W4385801322C33923547 @default.
• W4385801322 hasConceptScore W4385801322C41008148 @default.
• W4385801322 hasConceptScore W4385801322C535046627 @default.
• W4385801322 hasConceptScore W4385801322C71924100 @default.
• W4385801322 hasConceptScore W4385801322C86803240 @default.
• W4385801322 hasLocation W43858013221 @default.
• W4385801322 hasOpenAccess W4385801322 @default.
• W4385801322 hasPrimaryLocation W43858013221 @default.
• W4385801322 hasRelatedWork W2350459918 @default.
• W4385801322 hasRelatedWork W2885094885 @default.
• W4385801322 hasRelatedWork W2961085424 @default.
• W4385801322 hasRelatedWork W3046775127 @default.
• W4385801322 hasRelatedWork W3170094116 @default.
• W4385801322 hasRelatedWork W4285260836 @default.
• W4385801322 hasRelatedWork W4286629047 @default.
• W4385801322 hasRelatedWork W4306321456 @default.
• W4385801322 hasRelatedWork W4306674287 @default.
• W4385801322 hasRelatedWork W4224009465 @default.
• W4385801322 isParatext "false" @default.
• W4385801322 isRetracted "false" @default.
• W4385801322 workType "article" @default.