Matches in SemOpenAlex for { <https://semopenalex.org/work/W4312416807> ?p ?o ?g. }
Showing items 1 to 86 of
86
with 100 items per page.
- W4312416807 abstract "In visual retrieval systems, updating the embedding model requires recomputing features for every piece of data. This expensive process is referred to as backfilling. Re-cently, the idea of backward compatible training (BCT) was proposed. To avoid the cost of backfilling, BCT modifies training of the new model to make its representations com-patible with those of the old model. However, BCT can sig-nificantly hinder the performance of the new model. In this work, we propose a new learning paradigm for representation learning: forward compatible training (FCT). In Fct, when the old model is trained, we also prepare for a future unknown version of the model. We propose learning side- information, an auxiliary feature for each sample which fa-cilitates future updates of the model. To develop a powerful and flexible frameworkfor model compatibility, we combine side-information with a forward transformation from old to newembeddings. Training of the new model is not modified, hence, its accuracy is not degraded. We demonstrate sig-nificant retrieval accuracy improvement compared to BCT for various datasets: ImageNet-1k (+18.1%), Places-365 (+5.4%), and VGG-Face2 (+8.3%). Fctobtains model compatibility when the new and old models are trained across different datasets, losses, and architectures. <sup xmlns:mml=http://www.w3.org/1998/Math/MathML xmlns:xlink=http://www.w3.org/1999/xlink>1</sup> <sup xmlns:mml=http://www.w3.org/1998/Math/MathML xmlns:xlink=http://www.w3.org/1999/xlink>1</sup> Code available at https://github.com/apple/ml-fct." @default.
- W4312416807 created "2023-01-04" @default.
- W4312416807 creator A5013120724 @default.
- W4312416807 creator A5026171268 @default.
- W4312416807 creator A5028613002 @default.
- W4312416807 creator A5059295598 @default.
- W4312416807 creator A5060895155 @default.
- W4312416807 date "2022-06-01" @default.
- W4312416807 modified "2023-09-26" @default.
- W4312416807 title "Forward Compatible Training for Large-Scale Embedding Retrieval Systems" @default.
- W4312416807 cites W1992454046 @default.
- W4312416807 cites W1999818274 @default.
- W4312416807 cites W2117539524 @default.
- W4312416807 cites W2145215286 @default.
- W4312416807 cites W2165698076 @default.
- W4312416807 cites W2194775991 @default.
- W4312416807 cites W2788388592 @default.
- W4312416807 cites W2963163009 @default.
- W4312416807 cites W2964189064 @default.
- W4312416807 cites W2969985801 @default.
- W4312416807 cites W2975127015 @default.
- W4312416807 cites W3034594226 @default.
- W4312416807 cites W3104235057 @default.
- W4312416807 cites W3159481202 @default.
- W4312416807 cites W4214521899 @default.
- W4312416807 doi "https://doi.org/10.1109/cvpr52688.2022.01878" @default.
- W4312416807 hasPublicationYear "2022" @default.
- W4312416807 type Work @default.
- W4312416807 citedByCount "3" @default.
- W4312416807 countsByYear W43124168072023 @default.
- W4312416807 crossrefType "proceedings-article" @default.
- W4312416807 hasAuthorship W4312416807A5013120724 @default.
- W4312416807 hasAuthorship W4312416807A5026171268 @default.
- W4312416807 hasAuthorship W4312416807A5028613002 @default.
- W4312416807 hasAuthorship W4312416807A5059295598 @default.
- W4312416807 hasAuthorship W4312416807A5060895155 @default.
- W4312416807 hasBestOaLocation W43124168072 @default.
- W4312416807 hasConcept C108583219 @default.
- W4312416807 hasConcept C119857082 @default.
- W4312416807 hasConcept C127313418 @default.
- W4312416807 hasConcept C154945302 @default.
- W4312416807 hasConcept C17409809 @default.
- W4312416807 hasConcept C177264268 @default.
- W4312416807 hasConcept C199360897 @default.
- W4312416807 hasConcept C204321447 @default.
- W4312416807 hasConcept C23123220 @default.
- W4312416807 hasConcept C2776760102 @default.
- W4312416807 hasConcept C2778648169 @default.
- W4312416807 hasConcept C2778827112 @default.
- W4312416807 hasConcept C41008148 @default.
- W4312416807 hasConcept C41608201 @default.
- W4312416807 hasConcept C43126263 @default.
- W4312416807 hasConcept C51632099 @default.
- W4312416807 hasConceptScore W4312416807C108583219 @default.
- W4312416807 hasConceptScore W4312416807C119857082 @default.
- W4312416807 hasConceptScore W4312416807C127313418 @default.
- W4312416807 hasConceptScore W4312416807C154945302 @default.
- W4312416807 hasConceptScore W4312416807C17409809 @default.
- W4312416807 hasConceptScore W4312416807C177264268 @default.
- W4312416807 hasConceptScore W4312416807C199360897 @default.
- W4312416807 hasConceptScore W4312416807C204321447 @default.
- W4312416807 hasConceptScore W4312416807C23123220 @default.
- W4312416807 hasConceptScore W4312416807C2776760102 @default.
- W4312416807 hasConceptScore W4312416807C2778648169 @default.
- W4312416807 hasConceptScore W4312416807C2778827112 @default.
- W4312416807 hasConceptScore W4312416807C41008148 @default.
- W4312416807 hasConceptScore W4312416807C41608201 @default.
- W4312416807 hasConceptScore W4312416807C43126263 @default.
- W4312416807 hasConceptScore W4312416807C51632099 @default.
- W4312416807 hasLocation W43124168071 @default.
- W4312416807 hasLocation W43124168072 @default.
- W4312416807 hasOpenAccess W4312416807 @default.
- W4312416807 hasPrimaryLocation W43124168071 @default.
- W4312416807 hasRelatedWork W2773616286 @default.
- W4312416807 hasRelatedWork W2911455822 @default.
- W4312416807 hasRelatedWork W2942293597 @default.
- W4312416807 hasRelatedWork W2942650110 @default.
- W4312416807 hasRelatedWork W2968586400 @default.
- W4312416807 hasRelatedWork W3021430260 @default.
- W4312416807 hasRelatedWork W3160244858 @default.
- W4312416807 hasRelatedWork W3200179079 @default.
- W4312416807 hasRelatedWork W4281986673 @default.
- W4312416807 hasRelatedWork W4309938360 @default.
- W4312416807 isParatext "false" @default.
- W4312416807 isRetracted "false" @default.
- W4312416807 workType "article" @default.