FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W4386075908> ?p ?o ?g. }

• W4386075908 abstract "Vision-language models trained with contrastive learning on large-scale noisy data are becoming increasingly popular for zero-shot recognition problems. In this paper we improve the following three aspects of the contrastive pre-training pipeline: dataset noise, model initialization and the training objective. First, we propose a straightforward filtering strategy titled Complexity, Action, and Text-spotting (CAT) that significantly reduces dataset size, while achieving improved performance across zero-shot vision-language tasks. Next, we propose an approach titled Concept Distillation to leverage strong unimodal representations for contrastive training that does not increase training complexity while outperforming prior work. Finally, we modify the traditional contrastive alignment objective, and propose an importance-sampling approach to up-sample the importance of hard-negatives without adding additional complexity. On an extensive zero-shot benchmark of 29 tasks, our Distilled and Hard-negative Training (DiHT) approach improves on 20 tasks compared to the baseline. Furthermore, for few-shot linear probing, we propose a novel approach that bridges the gap between zero-shot and few-shot performance, substantially improving over prior work. Models are available at github.com/facebookresearch/diht." @default.
• W4386075908 created "2023-08-23" @default.
• W4386075908 creator A5006043629 @default.
• W4386075908 creator A5015469961 @default.
• W4386075908 creator A5020152678 @default.
• W4386075908 creator A5038939197 @default.
• W4386075908 creator A5052408504 @default.
• W4386075908 creator A5057748562 @default.
• W4386075908 creator A5068664439 @default.
• W4386075908 creator A5076888060 @default.
• W4386075908 creator A5077105480 @default.
• W4386075908 date "2023-06-01" @default.
• W4386075908 modified "2023-09-27" @default.
• W4386075908 title "Filtering, Distillation, and Hard Negatives for Vision-Language Pre-Training" @default.
• W4386075908 cites W1773149199 @default.
• W4386075908 cites W1977295328 @default.
• W4386075908 cites W1977766639 @default.
• W4386075908 cites W2031489346 @default.
• W4386075908 cites W2047643928 @default.
• W4386075908 cites W2117539524 @default.
• W4386075908 cites W2138011018 @default.
• W4386075908 cites W2277195237 @default.
• W4386075908 cites W2294370754 @default.
• W4386075908 cites W2533598788 @default.
• W4386075908 cites W2560647685 @default.
• W4386075908 cites W2560730294 @default.
• W4386075908 cites W2592335154 @default.
• W4386075908 cites W2904170036 @default.
• W4386075908 cites W2983128379 @default.
• W4386075908 cites W3034342078 @default.
• W4386075908 cites W3034781633 @default.
• W4386075908 cites W3035160371 @default.
• W4386075908 cites W3035524453 @default.
• W4386075908 cites W3138516171 @default.
• W4386075908 cites W3171007011 @default.
• W4386075908 cites W3205961173 @default.
• W4386075908 cites W3213454282 @default.
• W4386075908 cites W4250589301 @default.
• W4386075908 cites W4288083516 @default.
• W4386075908 cites W4312261477 @default.
• W4386075908 cites W4312480718 @default.
• W4386075908 cites W4312626422 @default.
• W4386075908 cites W4312777209 @default.
• W4386075908 cites W4312877428 @default.
• W4386075908 cites W4312910992 @default.
• W4386075908 cites W4313064375 @default.
• W4386075908 cites W4313136445 @default.
• W4386075908 doi "https://doi.org/10.1109/cvpr52729.2023.00673" @default.
• W4386075908 hasPublicationYear "2023" @default.
• W4386075908 type Work @default.
• W4386075908 citedByCount "0" @default.
• W4386075908 crossrefType "proceedings-article" @default.
• W4386075908 hasAuthorship W4386075908A5006043629 @default.
• W4386075908 hasAuthorship W4386075908A5015469961 @default.
• W4386075908 hasAuthorship W4386075908A5020152678 @default.
• W4386075908 hasAuthorship W4386075908A5038939197 @default.
• W4386075908 hasAuthorship W4386075908A5052408504 @default.
• W4386075908 hasAuthorship W4386075908A5057748562 @default.
• W4386075908 hasAuthorship W4386075908A5068664439 @default.
• W4386075908 hasAuthorship W4386075908A5076888060 @default.
• W4386075908 hasAuthorship W4386075908A5077105480 @default.
• W4386075908 hasConcept C114466953 @default.
• W4386075908 hasConcept C119857082 @default.
• W4386075908 hasConcept C13280743 @default.
• W4386075908 hasConcept C153083717 @default.
• W4386075908 hasConcept C154945302 @default.
• W4386075908 hasConcept C185798385 @default.
• W4386075908 hasConcept C199360897 @default.
• W4386075908 hasConcept C204321447 @default.
• W4386075908 hasConcept C205649164 @default.
• W4386075908 hasConcept C28490314 @default.
• W4386075908 hasConcept C41008148 @default.
• W4386075908 hasConcept C43521106 @default.
• W4386075908 hasConceptScore W4386075908C114466953 @default.
• W4386075908 hasConceptScore W4386075908C119857082 @default.
• W4386075908 hasConceptScore W4386075908C13280743 @default.
• W4386075908 hasConceptScore W4386075908C153083717 @default.
• W4386075908 hasConceptScore W4386075908C154945302 @default.
• W4386075908 hasConceptScore W4386075908C185798385 @default.
• W4386075908 hasConceptScore W4386075908C199360897 @default.
• W4386075908 hasConceptScore W4386075908C204321447 @default.
• W4386075908 hasConceptScore W4386075908C205649164 @default.
• W4386075908 hasConceptScore W4386075908C28490314 @default.
• W4386075908 hasConceptScore W4386075908C41008148 @default.
• W4386075908 hasConceptScore W4386075908C43521106 @default.
• W4386075908 hasLocation W43860759081 @default.
• W4386075908 hasOpenAccess W4386075908 @default.
• W4386075908 hasPrimaryLocation W43860759081 @default.
• W4386075908 hasRelatedWork W112744582 @default.
• W4386075908 hasRelatedWork W1485630101 @default.
• W4386075908 hasRelatedWork W2020540721 @default.
• W4386075908 hasRelatedWork W2275429949 @default.
• W4386075908 hasRelatedWork W2374442885 @default.
• W4386075908 hasRelatedWork W2374512474 @default.
• W4386075908 hasRelatedWork W2498017833 @default.
• W4386075908 hasRelatedWork W2961085424 @default.
• W4386075908 hasRelatedWork W3123920941 @default.
• W4386075908 hasRelatedWork W4306674287 @default.
• W4386075908 isParatext "false" @default.
• W4386075908 isRetracted "false" @default.

• next