Matches in SemOpenAlex for { <https://semopenalex.org/work/W3016092023> ?p ?o ?g. }
- W3016092023 abstract "Despite advances in neural language modeling, obtaining a good model on a large scale multi-domain dataset still remains a difficult task. We propose training methods for building neural language models for such a task, which are not only domain robust, but reasonable in model size and fast for evaluation. We combine knowledge distillation from pretrained domain expert language models with the noise contrastive estimation (NCE) loss. Knowledge distillation allows to train a single student model which is both compact and domain robust, while the use of NCE loss makes the model self-normalized, which enables fast evaluation. We conduct experiments on a large English multi-domain speech recognition dataset provided by AppTek. The resulting student model is of the size of one domain expert, while it gives similar perplexities as various teacher models on their expert domain; the model is self-normalized, allowing for 30% faster first pass decoding than the naive models which require the full soft- max computation, and finally it gives improvements of more than 8% relative in terms of word error rate over a large multidomain 4-gram count model trained on more than 10 B words." @default.
- W3016092023 created "2020-04-17" @default.
- W3016092023 creator A5002810304 @default.
- W3016092023 creator A5010929553 @default.
- W3016092023 creator A5015194760 @default.
- W3016092023 creator A5085516000 @default.
- W3016092023 creator A5087367411 @default.
- W3016092023 date "2020-05-01" @default.
- W3016092023 modified "2023-10-14" @default.
- W3016092023 title "Domain Robust, Fast, and Compact Neural Language Models" @default.
- W3016092023 cites W1520465330 @default.
- W3016092023 cites W179875071 @default.
- W3016092023 cites W1934041838 @default.
- W3016092023 cites W2024200390 @default.
- W3016092023 cites W2058641082 @default.
- W3016092023 cites W2064675550 @default.
- W3016092023 cites W2100664567 @default.
- W3016092023 cites W2158069733 @default.
- W3016092023 cites W2251682575 @default.
- W3016092023 cites W2294370754 @default.
- W3016092023 cites W2296545762 @default.
- W3016092023 cites W2402268235 @default.
- W3016092023 cites W2799923439 @default.
- W3016092023 cites W2889181648 @default.
- W3016092023 cites W2890487780 @default.
- W3016092023 cites W2891628540 @default.
- W3016092023 cites W2936295667 @default.
- W3016092023 cites W2938110959 @default.
- W3016092023 cites W2943845043 @default.
- W3016092023 cites W2972320704 @default.
- W3016092023 cites W2972399846 @default.
- W3016092023 cites W2973215447 @default.
- W3016092023 cites W2890012642 @default.
- W3016092023 doi "https://doi.org/10.1109/icassp40776.2020.9054399" @default.
- W3016092023 hasPublicationYear "2020" @default.
- W3016092023 type Work @default.
- W3016092023 sameAs 3016092023 @default.
- W3016092023 citedByCount "3" @default.
- W3016092023 countsByYear W30160920232020 @default.
- W3016092023 countsByYear W30160920232022 @default.
- W3016092023 crossrefType "proceedings-article" @default.
- W3016092023 hasAuthorship W3016092023A5002810304 @default.
- W3016092023 hasAuthorship W3016092023A5010929553 @default.
- W3016092023 hasAuthorship W3016092023A5015194760 @default.
- W3016092023 hasAuthorship W3016092023A5085516000 @default.
- W3016092023 hasAuthorship W3016092023A5087367411 @default.
- W3016092023 hasConcept C11413529 @default.
- W3016092023 hasConcept C115961682 @default.
- W3016092023 hasConcept C119857082 @default.
- W3016092023 hasConcept C134306372 @default.
- W3016092023 hasConcept C137293760 @default.
- W3016092023 hasConcept C154945302 @default.
- W3016092023 hasConcept C162324750 @default.
- W3016092023 hasConcept C187736073 @default.
- W3016092023 hasConcept C204321447 @default.
- W3016092023 hasConcept C207685749 @default.
- W3016092023 hasConcept C2524010 @default.
- W3016092023 hasConcept C2780451532 @default.
- W3016092023 hasConcept C28490314 @default.
- W3016092023 hasConcept C33923547 @default.
- W3016092023 hasConcept C36503486 @default.
- W3016092023 hasConcept C41008148 @default.
- W3016092023 hasConcept C45374587 @default.
- W3016092023 hasConcept C57273362 @default.
- W3016092023 hasConcept C90805587 @default.
- W3016092023 hasConcept C92548554 @default.
- W3016092023 hasConcept C99498987 @default.
- W3016092023 hasConceptScore W3016092023C11413529 @default.
- W3016092023 hasConceptScore W3016092023C115961682 @default.
- W3016092023 hasConceptScore W3016092023C119857082 @default.
- W3016092023 hasConceptScore W3016092023C134306372 @default.
- W3016092023 hasConceptScore W3016092023C137293760 @default.
- W3016092023 hasConceptScore W3016092023C154945302 @default.
- W3016092023 hasConceptScore W3016092023C162324750 @default.
- W3016092023 hasConceptScore W3016092023C187736073 @default.
- W3016092023 hasConceptScore W3016092023C204321447 @default.
- W3016092023 hasConceptScore W3016092023C207685749 @default.
- W3016092023 hasConceptScore W3016092023C2524010 @default.
- W3016092023 hasConceptScore W3016092023C2780451532 @default.
- W3016092023 hasConceptScore W3016092023C28490314 @default.
- W3016092023 hasConceptScore W3016092023C33923547 @default.
- W3016092023 hasConceptScore W3016092023C36503486 @default.
- W3016092023 hasConceptScore W3016092023C41008148 @default.
- W3016092023 hasConceptScore W3016092023C45374587 @default.
- W3016092023 hasConceptScore W3016092023C57273362 @default.
- W3016092023 hasConceptScore W3016092023C90805587 @default.
- W3016092023 hasConceptScore W3016092023C92548554 @default.
- W3016092023 hasConceptScore W3016092023C99498987 @default.
- W3016092023 hasLocation W30160920231 @default.
- W3016092023 hasOpenAccess W3016092023 @default.
- W3016092023 hasPrimaryLocation W30160920231 @default.
- W3016092023 hasRelatedWork W1539050421 @default.
- W3016092023 hasRelatedWork W1563147278 @default.
- W3016092023 hasRelatedWork W2359001871 @default.
- W3016092023 hasRelatedWork W2374471852 @default.
- W3016092023 hasRelatedWork W2887872604 @default.
- W3016092023 hasRelatedWork W3093097038 @default.
- W3016092023 hasRelatedWork W4205820553 @default.
- W3016092023 hasRelatedWork W4312304159 @default.
- W3016092023 hasRelatedWork W4385571594 @default.