Matches in SemOpenAlex for { <https://semopenalex.org/work/W4385571295> ?p ?o ?g. }
Showing items 1 to 61 of
61
with 100 items per page.
- W4385571295 abstract "In this work, we argue that non-autoregressive (NAR) sequence generative models can equivalently be regarded as an iterative refinement process towards the target sequence, implying an underlying dynamical system of NAR model: z = f (z, x) → y. In such a way, the optimal prediction of a NAR model should be the equilibrium state of its dynamics if given infinitely many iterations. However, this is infeasible in practice due to limited computational and memory budgets. To this end, we propose DEQNAR to directly solve for the equilibrium state of NAR models based on deep equilibrium networks (Bai et al., 2019) with black-box root-finding solvers and back-propagate through the equilibrium point via implicit differentiation with constant memory. We conduct extensive experiments on four WMT machine translation benchmarks. Our main findings show that DEQNAR can indeed converge to a more accurate prediction and is a general-purpose framework that consistently helps yield substantial improvement for several strong NAR backbones." @default.
- W4385571295 created "2023-08-05" @default.
- W4385571295 creator A5033423455 @default.
- W4385571295 creator A5049341927 @default.
- W4385571295 creator A5077749470 @default.
- W4385571295 date "2023-01-01" @default.
- W4385571295 modified "2023-10-18" @default.
- W4385571295 title "Deep Equilibrium Non-Autoregressive Sequence Learning" @default.
- W4385571295 doi "https://doi.org/10.18653/v1/2023.findings-acl.747" @default.
- W4385571295 hasPublicationYear "2023" @default.
- W4385571295 type Work @default.
- W4385571295 citedByCount "0" @default.
- W4385571295 crossrefType "proceedings-article" @default.
- W4385571295 hasAuthorship W4385571295A5033423455 @default.
- W4385571295 hasAuthorship W4385571295A5049341927 @default.
- W4385571295 hasAuthorship W4385571295A5077749470 @default.
- W4385571295 hasBestOaLocation W43855712951 @default.
- W4385571295 hasConcept C11413529 @default.
- W4385571295 hasConcept C126255220 @default.
- W4385571295 hasConcept C134306372 @default.
- W4385571295 hasConcept C149782125 @default.
- W4385571295 hasConcept C159877910 @default.
- W4385571295 hasConcept C2778112365 @default.
- W4385571295 hasConcept C28826006 @default.
- W4385571295 hasConcept C33923547 @default.
- W4385571295 hasConcept C41008148 @default.
- W4385571295 hasConcept C48103436 @default.
- W4385571295 hasConcept C54355233 @default.
- W4385571295 hasConcept C78045399 @default.
- W4385571295 hasConcept C86803240 @default.
- W4385571295 hasConcept C94766913 @default.
- W4385571295 hasConceptScore W4385571295C11413529 @default.
- W4385571295 hasConceptScore W4385571295C126255220 @default.
- W4385571295 hasConceptScore W4385571295C134306372 @default.
- W4385571295 hasConceptScore W4385571295C149782125 @default.
- W4385571295 hasConceptScore W4385571295C159877910 @default.
- W4385571295 hasConceptScore W4385571295C2778112365 @default.
- W4385571295 hasConceptScore W4385571295C28826006 @default.
- W4385571295 hasConceptScore W4385571295C33923547 @default.
- W4385571295 hasConceptScore W4385571295C41008148 @default.
- W4385571295 hasConceptScore W4385571295C48103436 @default.
- W4385571295 hasConceptScore W4385571295C54355233 @default.
- W4385571295 hasConceptScore W4385571295C78045399 @default.
- W4385571295 hasConceptScore W4385571295C86803240 @default.
- W4385571295 hasConceptScore W4385571295C94766913 @default.
- W4385571295 hasLocation W43855712951 @default.
- W4385571295 hasOpenAccess W4385571295 @default.
- W4385571295 hasPrimaryLocation W43855712951 @default.
- W4385571295 hasRelatedWork W1995849744 @default.
- W4385571295 hasRelatedWork W2028868645 @default.
- W4385571295 hasRelatedWork W2057419801 @default.
- W4385571295 hasRelatedWork W2057898520 @default.
- W4385571295 hasRelatedWork W2079942648 @default.
- W4385571295 hasRelatedWork W2153859070 @default.
- W4385571295 hasRelatedWork W2160749004 @default.
- W4385571295 hasRelatedWork W2386767533 @default.
- W4385571295 hasRelatedWork W1543105847 @default.
- W4385571295 hasRelatedWork W2069162488 @default.
- W4385571295 isParatext "false" @default.
- W4385571295 isRetracted "false" @default.
- W4385571295 workType "article" @default.