Matches in SemOpenAlex for { <https://semopenalex.org/work/W3085939014> ?p ?o ?g. }
- W3085939014 abstract "We propose an efficient inference procedure for non-autoregressive machine translation that iteratively refines translation purely in the continuous space. Given a continuous latent variable model for machine translation (Shu et al., 2020), we train an inference network to approximate the gradient of the marginal log probability of the target sentence, using only the latent variable as input. This allows us to use gradient-based optimization to find the target sentence at inference time that approximately maximizes its marginal probability. As each refinement step only involves computation in the latent space of low dimensionality (we use 8 in our experiments), we avoid computational overhead incurred by existing non-autoregressive inference procedures that often refine in token space. We compare our approach to a recently proposed EM-like inference procedure (Shu et al., 2020) that optimizes in a hybrid space, consisting of both discrete and continuous variables. We evaluate our approach on WMT'14 En-De, WMT'16 Ro-En and IWSLT'16 De-En, and observe two advantages over the EM-like inference: (1) it is computationally efficient, i.e. each refinement step is twice as fast, and (2) it is more effective, resulting in higher marginal probabilities and BLEU scores with the same number of refinement steps. On WMT'14 En-De, for instance, our approach is able to decode 6.2 times faster than the autoregressive model with minimal degradation to translation quality (0.9 BLEU)." @default.
- W3085939014 created "2020-09-21" @default.
- W3085939014 creator A5028843482 @default.
- W3085939014 creator A5030087918 @default.
- W3085939014 creator A5077805110 @default.
- W3085939014 date "2020-09-15" @default.
- W3085939014 modified "2023-10-03" @default.
- W3085939014 title "Iterative Refinement in the Continuous Space for Non-Autoregressive Neural Machine Translation" @default.
- W3085939014 cites W1505878979 @default.
- W3085939014 cites W1802356529 @default.
- W3085939014 cites W2013035813 @default.
- W3085939014 cites W2120340025 @default.
- W3085939014 cites W2120513984 @default.
- W3085939014 cites W2161914416 @default.
- W3085939014 cites W2539809671 @default.
- W3085939014 cites W2540413092 @default.
- W3085939014 cites W2547875792 @default.
- W3085939014 cites W2548228487 @default.
- W3085939014 cites W2553718801 @default.
- W3085939014 cites W2614634292 @default.
- W3085939014 cites W2738371943 @default.
- W3085939014 cites W2740094762 @default.
- W3085939014 cites W2769196856 @default.
- W3085939014 cites W2788330850 @default.
- W3085939014 cites W2789543585 @default.
- W3085939014 cites W2804821933 @default.
- W3085939014 cites W2885421725 @default.
- W3085939014 cites W2890397703 @default.
- W3085939014 cites W2933138175 @default.
- W3085939014 cites W2947898088 @default.
- W3085939014 cites W2948629866 @default.
- W3085939014 cites W2951730630 @default.
- W3085939014 cites W2962801832 @default.
- W3085939014 cites W2963250244 @default.
- W3085939014 cites W2963403868 @default.
- W3085939014 cites W2963434219 @default.
- W3085939014 cites W2963532001 @default.
- W3085939014 cites W2964076774 @default.
- W3085939014 cites W2964121744 @default.
- W3085939014 cites W2964205912 @default.
- W3085939014 cites W2971034910 @default.
- W3085939014 cites W2985694911 @default.
- W3085939014 cites W2988975212 @default.
- W3085939014 cites W2996843693 @default.
- W3085939014 cites W3006365256 @default.
- W3085939014 cites W3022372506 @default.
- W3085939014 cites W3098269892 @default.
- W3085939014 doi "https://doi.org/10.48550/arxiv.2009.07177" @default.
- W3085939014 hasPublicationYear "2020" @default.
- W3085939014 type Work @default.
- W3085939014 sameAs 3085939014 @default.
- W3085939014 citedByCount "1" @default.
- W3085939014 countsByYear W30859390142019 @default.
- W3085939014 crossrefType "posted-content" @default.
- W3085939014 hasAuthorship W3085939014A5028843482 @default.
- W3085939014 hasAuthorship W3085939014A5030087918 @default.
- W3085939014 hasAuthorship W3085939014A5077805110 @default.
- W3085939014 hasBestOaLocation W30859390141 @default.
- W3085939014 hasConcept C104317684 @default.
- W3085939014 hasConcept C105580179 @default.
- W3085939014 hasConcept C105795698 @default.
- W3085939014 hasConcept C11413529 @default.
- W3085939014 hasConcept C149364088 @default.
- W3085939014 hasConcept C154945302 @default.
- W3085939014 hasConcept C159877910 @default.
- W3085939014 hasConcept C185592680 @default.
- W3085939014 hasConcept C203005215 @default.
- W3085939014 hasConcept C2776214188 @default.
- W3085939014 hasConcept C33923547 @default.
- W3085939014 hasConcept C41008148 @default.
- W3085939014 hasConcept C51167844 @default.
- W3085939014 hasConcept C55493867 @default.
- W3085939014 hasConceptScore W3085939014C104317684 @default.
- W3085939014 hasConceptScore W3085939014C105580179 @default.
- W3085939014 hasConceptScore W3085939014C105795698 @default.
- W3085939014 hasConceptScore W3085939014C11413529 @default.
- W3085939014 hasConceptScore W3085939014C149364088 @default.
- W3085939014 hasConceptScore W3085939014C154945302 @default.
- W3085939014 hasConceptScore W3085939014C159877910 @default.
- W3085939014 hasConceptScore W3085939014C185592680 @default.
- W3085939014 hasConceptScore W3085939014C203005215 @default.
- W3085939014 hasConceptScore W3085939014C2776214188 @default.
- W3085939014 hasConceptScore W3085939014C33923547 @default.
- W3085939014 hasConceptScore W3085939014C41008148 @default.
- W3085939014 hasConceptScore W3085939014C51167844 @default.
- W3085939014 hasConceptScore W3085939014C55493867 @default.
- W3085939014 hasLocation W30859390141 @default.
- W3085939014 hasOpenAccess W3085939014 @default.
- W3085939014 hasPrimaryLocation W30859390141 @default.
- W3085939014 hasRelatedWork W10455605 @default.
- W3085939014 hasRelatedWork W10667231 @default.
- W3085939014 hasRelatedWork W11012074 @default.
- W3085939014 hasRelatedWork W11792228 @default.
- W3085939014 hasRelatedWork W12732426 @default.
- W3085939014 hasRelatedWork W13657349 @default.
- W3085939014 hasRelatedWork W13688497 @default.
- W3085939014 hasRelatedWork W1619002 @default.
- W3085939014 hasRelatedWork W3630569 @default.
- W3085939014 hasRelatedWork W7571534 @default.
- W3085939014 isParatext "false" @default.