Matches in SemOpenAlex for { <https://semopenalex.org/work/W3125261104> ?p ?o ?g. }
- W3125261104 abstract "We present a plug-in replacement for batch normalization (BN) called exponential moving average normalization (EMAN), which improves the performance of existing student-teacher based self- and semi-supervised learning techniques. Unlike the standard BN, where the statistics are computed within each batch, EMAN, used in the teacher, updates its statistics by exponential moving average from the BN statistics of the student. This design reduces the intrinsic cross-sample dependency of BN and enhances the generalization of the teacher. EMAN improves strong baselines for self-supervised learning by 4-6/1-2 points and semi-supervised learning by about 7/2 points, when 1%/10% supervised labels are available on ImageNet. These improvements are consistent across methods, network architectures, training duration, and datasets, demonstrating the general effectiveness of this technique. The code is available at https://github.com/amazon-research/exponential-moving-average-normalization." @default.
- W3125261104 created "2021-02-01" @default.
- W3125261104 creator A5001760915 @default.
- W3125261104 creator A5010126054 @default.
- W3125261104 creator A5038328783 @default.
- W3125261104 creator A5044978076 @default.
- W3125261104 creator A5052551454 @default.
- W3125261104 creator A5087812675 @default.
- W3125261104 date "2021-01-21" @default.
- W3125261104 modified "2023-09-27" @default.
- W3125261104 title "Exponential Moving Average Normalization for Self-supervised and Semi-supervised Learning" @default.
- W3125261104 cites W1821462560 @default.
- W3125261104 cites W1836465849 @default.
- W3125261104 cites W1861492603 @default.
- W3125261104 cites W2031489346 @default.
- W3125261104 cites W2095705004 @default.
- W3125261104 cites W2117539524 @default.
- W3125261104 cites W2119821739 @default.
- W3125261104 cites W2136504847 @default.
- W3125261104 cites W2138621090 @default.
- W3125261104 cites W2190691619 @default.
- W3125261104 cites W2194775991 @default.
- W3125261104 cites W2321533354 @default.
- W3125261104 cites W2326925005 @default.
- W3125261104 cites W2502312327 @default.
- W3125261104 cites W2592691248 @default.
- W3125261104 cites W2612983688 @default.
- W3125261104 cites W2795783309 @default.
- W3125261104 cites W2798991696 @default.
- W3125261104 cites W2909869271 @default.
- W3125261104 cites W2944828972 @default.
- W3125261104 cites W2946856970 @default.
- W3125261104 cites W2962369866 @default.
- W3125261104 cites W2962742544 @default.
- W3125261104 cites W2963173418 @default.
- W3125261104 cites W2963341956 @default.
- W3125261104 cites W2963397674 @default.
- W3125261104 cites W2964159205 @default.
- W3125261104 cites W2970971581 @default.
- W3125261104 cites W2981952041 @default.
- W3125261104 cites W2990873191 @default.
- W3125261104 cites W2995333370 @default.
- W3125261104 cites W2998388430 @default.
- W3125261104 cites W3001197829 @default.
- W3125261104 cites W3009561768 @default.
- W3125261104 cites W3022061250 @default.
- W3125261104 cites W3023371261 @default.
- W3125261104 cites W3034239448 @default.
- W3125261104 cites W3034781633 @default.
- W3125261104 cites W3034978746 @default.
- W3125261104 cites W3035058308 @default.
- W3125261104 cites W3035160371 @default.
- W3125261104 cites W3035524453 @default.
- W3125261104 cites W3095121901 @default.
- W3125261104 cites W3100859887 @default.
- W3125261104 cites W3101821705 @default.
- W3125261104 cites W3106031848 @default.
- W3125261104 cites W3108655343 @default.
- W3125261104 doi "https://doi.org/10.48550/arxiv.2101.08482" @default.
- W3125261104 hasPublicationYear "2021" @default.
- W3125261104 type Work @default.
- W3125261104 sameAs 3125261104 @default.
- W3125261104 citedByCount "2" @default.
- W3125261104 countsByYear W31252611042021 @default.
- W3125261104 crossrefType "posted-content" @default.
- W3125261104 hasAuthorship W3125261104A5001760915 @default.
- W3125261104 hasAuthorship W3125261104A5010126054 @default.
- W3125261104 hasAuthorship W3125261104A5038328783 @default.
- W3125261104 hasAuthorship W3125261104A5044978076 @default.
- W3125261104 hasAuthorship W3125261104A5052551454 @default.
- W3125261104 hasAuthorship W3125261104A5087812675 @default.
- W3125261104 hasBestOaLocation W31252611041 @default.
- W3125261104 hasConcept C119857082 @default.
- W3125261104 hasConcept C134306372 @default.
- W3125261104 hasConcept C136389625 @default.
- W3125261104 hasConcept C136886441 @default.
- W3125261104 hasConcept C144024400 @default.
- W3125261104 hasConcept C151376022 @default.
- W3125261104 hasConcept C154945302 @default.
- W3125261104 hasConcept C177148314 @default.
- W3125261104 hasConcept C19165224 @default.
- W3125261104 hasConcept C33923547 @default.
- W3125261104 hasConcept C41008148 @default.
- W3125261104 hasConcept C50644808 @default.
- W3125261104 hasConceptScore W3125261104C119857082 @default.
- W3125261104 hasConceptScore W3125261104C134306372 @default.
- W3125261104 hasConceptScore W3125261104C136389625 @default.
- W3125261104 hasConceptScore W3125261104C136886441 @default.
- W3125261104 hasConceptScore W3125261104C144024400 @default.
- W3125261104 hasConceptScore W3125261104C151376022 @default.
- W3125261104 hasConceptScore W3125261104C154945302 @default.
- W3125261104 hasConceptScore W3125261104C177148314 @default.
- W3125261104 hasConceptScore W3125261104C19165224 @default.
- W3125261104 hasConceptScore W3125261104C33923547 @default.
- W3125261104 hasConceptScore W3125261104C41008148 @default.
- W3125261104 hasConceptScore W3125261104C50644808 @default.
- W3125261104 hasLocation W31252611041 @default.
- W3125261104 hasOpenAccess W3125261104 @default.
- W3125261104 hasPrimaryLocation W31252611041 @default.
- W3125261104 hasRelatedWork W2981850339 @default.