Matches in SemOpenAlex for { <https://semopenalex.org/work/W3217573676> ?p ?o ?g. }
Showing items 1 to 84 of
84
with 100 items per page.
- W3217573676 abstract "Recent research suggests that systematic generalization in natural language understanding remains a challenge for state-of-the-art neural models such as Transformers and Graph Neural Networks. To tackle this challenge, we propose Edge Transformer, a new model that combines inspiration from Transformers and rule-based symbolic AI. The first key idea in Edge Transformers is to associate vector states with every edge, that is, with every pair of input nodes -- as opposed to just every node, as it is done in the Transformer model. The second major innovation is a triangular attention mechanism that updates edge representations in a way that is inspired by unification from logic programming. We evaluate Edge Transformer on compositional generalization benchmarks in relational reasoning, semantic parsing, and dependency parsing. In all three settings, the Edge Transformer outperforms Relation-aware, Universal and classical Transformer baselines." @default.
- W3217573676 created "2021-12-06" @default.
- W3217573676 creator A5010465328 @default.
- W3217573676 creator A5078977334 @default.
- W3217573676 creator A5089919672 @default.
- W3217573676 date "2021-12-01" @default.
- W3217573676 modified "2023-09-23" @default.
- W3217573676 title "Systematic Generalization with Edge Transformers" @default.
- W3217573676 cites W1501856433 @default.
- W3217573676 cites W2116341502 @default.
- W3217573676 cites W2519887557 @default.
- W3217573676 cites W2552110825 @default.
- W3217573676 cites W2593652440 @default.
- W3217573676 cites W2617995259 @default.
- W3217573676 cites W2766453196 @default.
- W3217573676 cites W2866343820 @default.
- W3217573676 cites W2908791737 @default.
- W3217573676 cites W2911109671 @default.
- W3217573676 cites W2913355189 @default.
- W3217573676 cites W2938432841 @default.
- W3217573676 cites W2950129230 @default.
- W3217573676 cites W2950407619 @default.
- W3217573676 cites W2952254971 @default.
- W3217573676 cites W2962924847 @default.
- W3217573676 cites W2963267799 @default.
- W3217573676 cites W2963341956 @default.
- W3217573676 cites W2963403868 @default.
- W3217573676 cites W2963925437 @default.
- W3217573676 cites W2964071299 @default.
- W3217573676 cites W2969219365 @default.
- W3217573676 cites W2970349885 @default.
- W3217573676 cites W2971107062 @default.
- W3217573676 cites W2996094825 @default.
- W3217573676 cites W3011798063 @default.
- W3217573676 cites W3016697633 @default.
- W3217573676 cites W3022131108 @default.
- W3217573676 cites W3037109418 @default.
- W3217573676 cites W3041492656 @default.
- W3217573676 cites W3043172396 @default.
- W3217573676 cites W3082274269 @default.
- W3217573676 cites W3091156754 @default.
- W3217573676 cites W3092252624 @default.
- W3217573676 cites W3093157543 @default.
- W3217573676 cites W3096609285 @default.
- W3217573676 cites W3101035550 @default.
- W3217573676 cites W3104739822 @default.
- W3217573676 cites W3110744880 @default.
- W3217573676 cites W3128306840 @default.
- W3217573676 cites W3191727383 @default.
- W3217573676 cites W3207306342 @default.
- W3217573676 cites W3197009789 @default.
- W3217573676 hasPublicationYear "2021" @default.
- W3217573676 type Work @default.
- W3217573676 sameAs 3217573676 @default.
- W3217573676 citedByCount "0" @default.
- W3217573676 crossrefType "posted-content" @default.
- W3217573676 hasAuthorship W3217573676A5010465328 @default.
- W3217573676 hasAuthorship W3217573676A5078977334 @default.
- W3217573676 hasAuthorship W3217573676A5089919672 @default.
- W3217573676 hasConcept C119599485 @default.
- W3217573676 hasConcept C127413603 @default.
- W3217573676 hasConcept C154945302 @default.
- W3217573676 hasConcept C165801399 @default.
- W3217573676 hasConcept C186644900 @default.
- W3217573676 hasConcept C199360897 @default.
- W3217573676 hasConcept C41008148 @default.
- W3217573676 hasConcept C66322947 @default.
- W3217573676 hasConcept C80444323 @default.
- W3217573676 hasConcept C96146094 @default.
- W3217573676 hasConceptScore W3217573676C119599485 @default.
- W3217573676 hasConceptScore W3217573676C127413603 @default.
- W3217573676 hasConceptScore W3217573676C154945302 @default.
- W3217573676 hasConceptScore W3217573676C165801399 @default.
- W3217573676 hasConceptScore W3217573676C186644900 @default.
- W3217573676 hasConceptScore W3217573676C199360897 @default.
- W3217573676 hasConceptScore W3217573676C41008148 @default.
- W3217573676 hasConceptScore W3217573676C66322947 @default.
- W3217573676 hasConceptScore W3217573676C80444323 @default.
- W3217573676 hasConceptScore W3217573676C96146094 @default.
- W3217573676 hasOpenAccess W3217573676 @default.
- W3217573676 isParatext "false" @default.
- W3217573676 isRetracted "false" @default.
- W3217573676 magId "3217573676" @default.
- W3217573676 workType "article" @default.