Matches in SemOpenAlex for { <https://semopenalex.org/work/W2890152612> ?p ?o ?g. }
Showing items 1 to 75 of
75
with 100 items per page.
- W2890152612 abstract "Deep autoregressive sequence-to-sequence models have demonstrated impressive performance across a wide variety of tasks in recent years. While common architecture classes such as recurrent, convolutional, and self-attention networks make different trade-offs between the amount of computation needed per layer and the length of the critical path at training time, generation still remains an inherently sequential process. To overcome this limitation, we propose a novel blockwise parallel decoding scheme in which we make predictions for multiple time steps in parallel then back off to the longest prefix validated by a scoring model. This allows for substantial theoretical improvements in generation speed when applied to architectures that can process output sequences in parallel. We verify our approach empirically through a series of experiments using state-of-the-art self-attention models for machine translation and image super-resolution, achieving iteration reductions of up to 2x over a baseline greedy decoder with no loss in quality, or up to 7x in exchange for a slight decrease in performance. In terms of wall-clock time, our fastest models exhibit real-time speedups of up to 4x over standard greedy decoding." @default.
- W2890152612 created "2018-09-27" @default.
- W2890152612 creator A5021878400 @default.
- W2890152612 creator A5022416424 @default.
- W2890152612 creator A5039013714 @default.
- W2890152612 date "2018-11-07" @default.
- W2890152612 modified "2023-09-27" @default.
- W2890152612 title "Blockwise Parallel Decoding for Deep Autoregressive Models" @default.
- W2890152612 hasPublicationYear "2018" @default.
- W2890152612 type Work @default.
- W2890152612 sameAs 2890152612 @default.
- W2890152612 citedByCount "0" @default.
- W2890152612 crossrefType "posted-content" @default.
- W2890152612 hasAuthorship W2890152612A5021878400 @default.
- W2890152612 hasAuthorship W2890152612A5022416424 @default.
- W2890152612 hasAuthorship W2890152612A5039013714 @default.
- W2890152612 hasConcept C111919701 @default.
- W2890152612 hasConcept C11413529 @default.
- W2890152612 hasConcept C149782125 @default.
- W2890152612 hasConcept C154945302 @default.
- W2890152612 hasConcept C159877910 @default.
- W2890152612 hasConcept C173608175 @default.
- W2890152612 hasConcept C199360897 @default.
- W2890152612 hasConcept C2777735758 @default.
- W2890152612 hasConcept C2778112365 @default.
- W2890152612 hasConcept C33923547 @default.
- W2890152612 hasConcept C41008148 @default.
- W2890152612 hasConcept C45374587 @default.
- W2890152612 hasConcept C54355233 @default.
- W2890152612 hasConcept C57273362 @default.
- W2890152612 hasConcept C86803240 @default.
- W2890152612 hasConcept C98045186 @default.
- W2890152612 hasConceptScore W2890152612C111919701 @default.
- W2890152612 hasConceptScore W2890152612C11413529 @default.
- W2890152612 hasConceptScore W2890152612C149782125 @default.
- W2890152612 hasConceptScore W2890152612C154945302 @default.
- W2890152612 hasConceptScore W2890152612C159877910 @default.
- W2890152612 hasConceptScore W2890152612C173608175 @default.
- W2890152612 hasConceptScore W2890152612C199360897 @default.
- W2890152612 hasConceptScore W2890152612C2777735758 @default.
- W2890152612 hasConceptScore W2890152612C2778112365 @default.
- W2890152612 hasConceptScore W2890152612C33923547 @default.
- W2890152612 hasConceptScore W2890152612C41008148 @default.
- W2890152612 hasConceptScore W2890152612C45374587 @default.
- W2890152612 hasConceptScore W2890152612C54355233 @default.
- W2890152612 hasConceptScore W2890152612C57273362 @default.
- W2890152612 hasConceptScore W2890152612C86803240 @default.
- W2890152612 hasConceptScore W2890152612C98045186 @default.
- W2890152612 hasLocation W28901526121 @default.
- W2890152612 hasOpenAccess W2890152612 @default.
- W2890152612 hasPrimaryLocation W28901526121 @default.
- W2890152612 hasRelatedWork W2036824485 @default.
- W2890152612 hasRelatedWork W2139034093 @default.
- W2890152612 hasRelatedWork W2758162718 @default.
- W2890152612 hasRelatedWork W2765402136 @default.
- W2890152612 hasRelatedWork W2806985991 @default.
- W2890152612 hasRelatedWork W2890770683 @default.
- W2890152612 hasRelatedWork W2891491149 @default.
- W2890152612 hasRelatedWork W2892799969 @default.
- W2890152612 hasRelatedWork W2914261901 @default.
- W2890152612 hasRelatedWork W2921917212 @default.
- W2890152612 hasRelatedWork W2960200274 @default.
- W2890152612 hasRelatedWork W2963946353 @default.
- W2890152612 hasRelatedWork W2964307104 @default.
- W2890152612 hasRelatedWork W2973163720 @default.
- W2890152612 hasRelatedWork W3026505846 @default.
- W2890152612 hasRelatedWork W3117464205 @default.
- W2890152612 hasRelatedWork W3132407408 @default.
- W2890152612 hasRelatedWork W3167633040 @default.
- W2890152612 hasRelatedWork W3172749677 @default.
- W2890152612 hasRelatedWork W3200312342 @default.
- W2890152612 isParatext "false" @default.
- W2890152612 isRetracted "false" @default.
- W2890152612 magId "2890152612" @default.
- W2890152612 workType "article" @default.