Matches in SemOpenAlex for { <https://semopenalex.org/work/W3213961506> ?p ?o ?g. }
Showing items 1 to 87 of
87
with 100 items per page.
- W3213961506 abstract "Adapters are light-weight modules that allow parameter-efficient fine-tuning of pretrained models. Specialized language and task adapters have recently been proposed to facilitate cross-lingual transfer of multilingual pretrained models (Pfeiffer et al., 2020b). However, this approach requires training a separate language adapter for every language one wishes to support, which can be impractical for languages with limited data. An intuitive solution is to use a related language adapter for the new language variety, but we observe that this solution can lead to sub-optimal performance. In this paper, we aim to improve the robustness of language adapters to uncovered languages without training new adapters. We find that ensembling multiple existing language adapters makes the fine-tuned model significantly more robust to other language varieties not included in these adapters. Building upon this observation, we propose Entropy Minimized Ensemble of Adapters (EMEA), a method that optimizes the ensemble weights of the pretrained language adapters for each test sentence by minimizing the entropy of its predictions. Experiments on three diverse groups of language varieties show that our method leads to significant improvements on both named entity recognition and part-of-speech tagging across all languages." @default.
- W3213961506 created "2021-11-22" @default.
- W3213961506 creator A5037310413 @default.
- W3213961506 creator A5040523461 @default.
- W3213961506 creator A5062910836 @default.
- W3213961506 creator A5068811427 @default.
- W3213961506 date "2021-09-10" @default.
- W3213961506 modified "2023-09-27" @default.
- W3213961506 title "Efficient Test Time Adapter Ensembling for Low-resource Language Varieties" @default.
- W3213961506 cites W1987093418 @default.
- W3213961506 cites W1995875735 @default.
- W3213961506 cites W2812315852 @default.
- W3213961506 cites W2915977242 @default.
- W3213961506 cites W2963211188 @default.
- W3213961506 cites W2963341956 @default.
- W3213961506 cites W2964303773 @default.
- W3213961506 cites W2970925270 @default.
- W3213961506 cites W2971207485 @default.
- W3213961506 cites W3004076319 @default.
- W3213961506 cites W3034408878 @default.
- W3213961506 cites W3035390927 @default.
- W3213961506 cites W3035579820 @default.
- W3213961506 cites W3099793224 @default.
- W3213961506 cites W3101498587 @default.
- W3213961506 cites W3105421296 @default.
- W3213961506 cites W3113057009 @default.
- W3213961506 cites W3117088788 @default.
- W3213961506 cites W3119438769 @default.
- W3213961506 cites W3119543788 @default.
- W3213961506 cites W3153597994 @default.
- W3213961506 cites W3153675281 @default.
- W3213961506 cites W3170666909 @default.
- W3213961506 cites W3173785189 @default.
- W3213961506 doi "https://doi.org/10.48550/arxiv.2109.04877" @default.
- W3213961506 hasPublicationYear "2021" @default.
- W3213961506 type Work @default.
- W3213961506 sameAs 3213961506 @default.
- W3213961506 citedByCount "0" @default.
- W3213961506 crossrefType "posted-content" @default.
- W3213961506 hasAuthorship W3213961506A5037310413 @default.
- W3213961506 hasAuthorship W3213961506A5040523461 @default.
- W3213961506 hasAuthorship W3213961506A5062910836 @default.
- W3213961506 hasAuthorship W3213961506A5068811427 @default.
- W3213961506 hasBestOaLocation W32139615061 @default.
- W3213961506 hasConcept C111919701 @default.
- W3213961506 hasConcept C136197465 @default.
- W3213961506 hasConcept C137293760 @default.
- W3213961506 hasConcept C138885662 @default.
- W3213961506 hasConcept C154945302 @default.
- W3213961506 hasConcept C177284502 @default.
- W3213961506 hasConcept C204321447 @default.
- W3213961506 hasConcept C2777530160 @default.
- W3213961506 hasConcept C28490314 @default.
- W3213961506 hasConcept C41008148 @default.
- W3213961506 hasConcept C41895202 @default.
- W3213961506 hasConcept C94922259 @default.
- W3213961506 hasConcept C9679016 @default.
- W3213961506 hasConceptScore W3213961506C111919701 @default.
- W3213961506 hasConceptScore W3213961506C136197465 @default.
- W3213961506 hasConceptScore W3213961506C137293760 @default.
- W3213961506 hasConceptScore W3213961506C138885662 @default.
- W3213961506 hasConceptScore W3213961506C154945302 @default.
- W3213961506 hasConceptScore W3213961506C177284502 @default.
- W3213961506 hasConceptScore W3213961506C204321447 @default.
- W3213961506 hasConceptScore W3213961506C2777530160 @default.
- W3213961506 hasConceptScore W3213961506C28490314 @default.
- W3213961506 hasConceptScore W3213961506C41008148 @default.
- W3213961506 hasConceptScore W3213961506C41895202 @default.
- W3213961506 hasConceptScore W3213961506C94922259 @default.
- W3213961506 hasConceptScore W3213961506C9679016 @default.
- W3213961506 hasLocation W32139615061 @default.
- W3213961506 hasOpenAccess W3213961506 @default.
- W3213961506 hasPrimaryLocation W32139615061 @default.
- W3213961506 hasRelatedWork W1517743118 @default.
- W3213961506 hasRelatedWork W1567338489 @default.
- W3213961506 hasRelatedWork W159132833 @default.
- W3213961506 hasRelatedWork W1978971213 @default.
- W3213961506 hasRelatedWork W2167717037 @default.
- W3213961506 hasRelatedWork W2952992889 @default.
- W3213961506 hasRelatedWork W3107474891 @default.
- W3213961506 hasRelatedWork W3200245893 @default.
- W3213961506 hasRelatedWork W3213961506 @default.
- W3213961506 hasRelatedWork W38394648 @default.
- W3213961506 isParatext "false" @default.
- W3213961506 isRetracted "false" @default.
- W3213961506 magId "3213961506" @default.
- W3213961506 workType "article" @default.