Matches in SemOpenAlex for { <https://semopenalex.org/work/W4372347449> ?p ?o ?g. }
Showing items 1 to 94 of
94
with 100 items per page.
- W4372347449 abstract "Self-supervised learning (SSL) models reshaped our approach to speech, language and vision. However their huge size and the opaque relations between their layers and tasks result in slow inference and network overthinking, where predictions made from the last layer of large models is worse than those made from intermediate layers. Early exit (EE) strategies can solve both issues by dynamically reducing computations at inference time for certain samples. Although popular for classification tasks in vision and language, EE has seen less use for sequence-to-sequence speech recognition (ASR) tasks where outputs from early layers are often degenerate. This challenge is further compounded when speech SSL models are applied on out-of-distribution (OOD) data. This paper first shows that SSL models do overthinking in ASR. We then motivate further research in EE by computing an optimal bound for performance versus speed trade-offs. To approach this bound we propose two new strategies for ASR: (1) we adapt the recently proposed patience strategy to ASR; and (2) we design a new EE strategy specific to ASR that performs better than all strategies previously introduced." @default.
- W4372347449 created "2023-05-07" @default.
- W4372347449 creator A5001291873 @default.
- W4372347449 creator A5021201726 @default.
- W4372347449 creator A5024324721 @default.
- W4372347449 date "2023-06-04" @default.
- W4372347449 modified "2023-09-27" @default.
- W4372347449 title "Avoid Overthinking in Self-Supervised Models for Speech Recognition" @default.
- W4372347449 cites W1494198834 @default.
- W4372347449 cites W2025713906 @default.
- W4372347449 cites W2043701535 @default.
- W4372347449 cites W2058094241 @default.
- W4372347449 cites W2127141656 @default.
- W4372347449 cites W2962677625 @default.
- W4372347449 cites W2962780374 @default.
- W4372347449 cites W2995181338 @default.
- W4372347449 cites W3015265920 @default.
- W4372347449 cites W3035038672 @default.
- W4372347449 cites W3154971029 @default.
- W4372347449 cites W3160106041 @default.
- W4372347449 cites W3162249256 @default.
- W4372347449 cites W3170113752 @default.
- W4372347449 cites W3177295825 @default.
- W4372347449 cites W3197580070 @default.
- W4372347449 cites W3198771897 @default.
- W4372347449 cites W3209059054 @default.
- W4372347449 cites W3209984917 @default.
- W4372347449 cites W4225285622 @default.
- W4372347449 cites W4226380987 @default.
- W4372347449 cites W4254751698 @default.
- W4372347449 cites W4312561350 @default.
- W4372347449 doi "https://doi.org/10.1109/icassp49357.2023.10095335" @default.
- W4372347449 hasPublicationYear "2023" @default.
- W4372347449 type Work @default.
- W4372347449 citedByCount "0" @default.
- W4372347449 crossrefType "proceedings-article" @default.
- W4372347449 hasAuthorship W4372347449A5001291873 @default.
- W4372347449 hasAuthorship W4372347449A5021201726 @default.
- W4372347449 hasAuthorship W4372347449A5024324721 @default.
- W4372347449 hasBestOaLocation W43723474491 @default.
- W4372347449 hasConcept C11413529 @default.
- W4372347449 hasConcept C119857082 @default.
- W4372347449 hasConcept C137293760 @default.
- W4372347449 hasConcept C154945302 @default.
- W4372347449 hasConcept C162324750 @default.
- W4372347449 hasConcept C178790620 @default.
- W4372347449 hasConcept C185592680 @default.
- W4372347449 hasConcept C187736073 @default.
- W4372347449 hasConcept C2776214188 @default.
- W4372347449 hasConcept C2778112365 @default.
- W4372347449 hasConcept C2779227376 @default.
- W4372347449 hasConcept C2780451532 @default.
- W4372347449 hasConcept C28490314 @default.
- W4372347449 hasConcept C35639132 @default.
- W4372347449 hasConcept C41008148 @default.
- W4372347449 hasConcept C45374587 @default.
- W4372347449 hasConcept C54355233 @default.
- W4372347449 hasConcept C86803240 @default.
- W4372347449 hasConceptScore W4372347449C11413529 @default.
- W4372347449 hasConceptScore W4372347449C119857082 @default.
- W4372347449 hasConceptScore W4372347449C137293760 @default.
- W4372347449 hasConceptScore W4372347449C154945302 @default.
- W4372347449 hasConceptScore W4372347449C162324750 @default.
- W4372347449 hasConceptScore W4372347449C178790620 @default.
- W4372347449 hasConceptScore W4372347449C185592680 @default.
- W4372347449 hasConceptScore W4372347449C187736073 @default.
- W4372347449 hasConceptScore W4372347449C2776214188 @default.
- W4372347449 hasConceptScore W4372347449C2778112365 @default.
- W4372347449 hasConceptScore W4372347449C2779227376 @default.
- W4372347449 hasConceptScore W4372347449C2780451532 @default.
- W4372347449 hasConceptScore W4372347449C28490314 @default.
- W4372347449 hasConceptScore W4372347449C35639132 @default.
- W4372347449 hasConceptScore W4372347449C41008148 @default.
- W4372347449 hasConceptScore W4372347449C45374587 @default.
- W4372347449 hasConceptScore W4372347449C54355233 @default.
- W4372347449 hasConceptScore W4372347449C86803240 @default.
- W4372347449 hasFunder F4320306076 @default.
- W4372347449 hasLocation W43723474491 @default.
- W4372347449 hasLocation W43723474492 @default.
- W4372347449 hasOpenAccess W4372347449 @default.
- W4372347449 hasPrimaryLocation W43723474491 @default.
- W4372347449 hasRelatedWork W2806424637 @default.
- W4372347449 hasRelatedWork W2834136616 @default.
- W4372347449 hasRelatedWork W2888289075 @default.
- W4372347449 hasRelatedWork W2947903144 @default.
- W4372347449 hasRelatedWork W2949367580 @default.
- W4372347449 hasRelatedWork W2961085424 @default.
- W4372347449 hasRelatedWork W3164092048 @default.
- W4372347449 hasRelatedWork W3204102400 @default.
- W4372347449 hasRelatedWork W4306674287 @default.
- W4372347449 hasRelatedWork W4224009465 @default.
- W4372347449 isParatext "false" @default.
- W4372347449 isRetracted "false" @default.
- W4372347449 workType "article" @default.