Matches in SemOpenAlex for { <https://semopenalex.org/work/W4281488380> ?p ?o ?g. }
Showing items 1 to 74 of
74
with 100 items per page.
- W4281488380 abstract "Multilingual machine translation suffers from negative interference across languages. A common solution is to relax parameter sharing with language-specific modules like adapters. However, adapters of related languages are unable to transfer information, and their total number of parameters becomes prohibitively expensive as the number of languages grows. In this work, we overcome these drawbacks using hyper-adapters -- hyper-networks that generate adapters from language and layer embeddings. While past work had poor results when scaling hyper-networks, we propose a rescaling fix that significantly improves convergence and enables training larger hyper-networks. We find that hyper-adapters are more parameter efficient than regular adapters, reaching the same performance with up to 12 times less parameters. When using the same number of parameters and FLOPS, our approach consistently outperforms regular adapters. Also, hyper-adapters converge faster than alternative approaches and scale better than regular dense networks. Our analysis shows that hyper-adapters learn to encode language relatedness, enabling positive transfer across languages." @default.
- W4281488380 created "2022-05-26" @default.
- W4281488380 creator A5023341622 @default.
- W4281488380 creator A5035665045 @default.
- W4281488380 creator A5065321401 @default.
- W4281488380 creator A5075668848 @default.
- W4281488380 date "2022-05-22" @default.
- W4281488380 modified "2023-10-10" @default.
- W4281488380 title "Multilingual Machine Translation with Hyper-Adapters" @default.
- W4281488380 doi "https://doi.org/10.48550/arxiv.2205.10835" @default.
- W4281488380 hasPublicationYear "2022" @default.
- W4281488380 type Work @default.
- W4281488380 citedByCount "0" @default.
- W4281488380 crossrefType "posted-content" @default.
- W4281488380 hasAuthorship W4281488380A5023341622 @default.
- W4281488380 hasAuthorship W4281488380A5035665045 @default.
- W4281488380 hasAuthorship W4281488380A5065321401 @default.
- W4281488380 hasAuthorship W4281488380A5075668848 @default.
- W4281488380 hasBestOaLocation W42814883801 @default.
- W4281488380 hasConcept C104317684 @default.
- W4281488380 hasConcept C105580179 @default.
- W4281488380 hasConcept C149364088 @default.
- W4281488380 hasConcept C154945302 @default.
- W4281488380 hasConcept C162324750 @default.
- W4281488380 hasConcept C173608175 @default.
- W4281488380 hasConcept C185592680 @default.
- W4281488380 hasConcept C203005215 @default.
- W4281488380 hasConcept C2524010 @default.
- W4281488380 hasConcept C2776175482 @default.
- W4281488380 hasConcept C2777303404 @default.
- W4281488380 hasConcept C33923547 @default.
- W4281488380 hasConcept C3826847 @default.
- W4281488380 hasConcept C41008148 @default.
- W4281488380 hasConcept C50522688 @default.
- W4281488380 hasConcept C55493867 @default.
- W4281488380 hasConcept C66746571 @default.
- W4281488380 hasConcept C80444323 @default.
- W4281488380 hasConcept C99844830 @default.
- W4281488380 hasConceptScore W4281488380C104317684 @default.
- W4281488380 hasConceptScore W4281488380C105580179 @default.
- W4281488380 hasConceptScore W4281488380C149364088 @default.
- W4281488380 hasConceptScore W4281488380C154945302 @default.
- W4281488380 hasConceptScore W4281488380C162324750 @default.
- W4281488380 hasConceptScore W4281488380C173608175 @default.
- W4281488380 hasConceptScore W4281488380C185592680 @default.
- W4281488380 hasConceptScore W4281488380C203005215 @default.
- W4281488380 hasConceptScore W4281488380C2524010 @default.
- W4281488380 hasConceptScore W4281488380C2776175482 @default.
- W4281488380 hasConceptScore W4281488380C2777303404 @default.
- W4281488380 hasConceptScore W4281488380C33923547 @default.
- W4281488380 hasConceptScore W4281488380C3826847 @default.
- W4281488380 hasConceptScore W4281488380C41008148 @default.
- W4281488380 hasConceptScore W4281488380C50522688 @default.
- W4281488380 hasConceptScore W4281488380C55493867 @default.
- W4281488380 hasConceptScore W4281488380C66746571 @default.
- W4281488380 hasConceptScore W4281488380C80444323 @default.
- W4281488380 hasConceptScore W4281488380C99844830 @default.
- W4281488380 hasLocation W42814883801 @default.
- W4281488380 hasLocation W42814883802 @default.
- W4281488380 hasOpenAccess W4281488380 @default.
- W4281488380 hasPrimaryLocation W42814883801 @default.
- W4281488380 hasRelatedWork W1989130879 @default.
- W4281488380 hasRelatedWork W2103419012 @default.
- W4281488380 hasRelatedWork W2354198838 @default.
- W4281488380 hasRelatedWork W2468279273 @default.
- W4281488380 hasRelatedWork W2988126442 @default.
- W4281488380 hasRelatedWork W3134206242 @default.
- W4281488380 hasRelatedWork W3205506801 @default.
- W4281488380 hasRelatedWork W4287272376 @default.
- W4281488380 hasRelatedWork W4315697128 @default.
- W4281488380 hasRelatedWork W4382323155 @default.
- W4281488380 isParatext "false" @default.
- W4281488380 isRetracted "false" @default.
- W4281488380 workType "article" @default.