Matches in SemOpenAlex for { <https://semopenalex.org/work/W4309869673> ?p ?o ?g. }
Showing items 1 to 61 of
61
with 100 items per page.
- W4309869673 abstract "While large pre-trained language models are powerful, their predictions often lack logical consistency across test inputs. For example, a state-of-the-art Macaw question-answering (QA) model answers 'Yes' to 'Is a sparrow a bird?' and 'Does a bird have feet?' but answers 'No' to 'Does a sparrow have feet?'. To address this failure mode, we propose a framework, Consistency Correction through Relation Detection, or ConCoRD, for boosting the consistency and accuracy of pre-trained NLP models using pre-trained natural language inference (NLI) models without fine-tuning or re-training. Given a batch of test inputs, ConCoRD samples several candidate outputs for each input and instantiates a factor graph that accounts for both the model's belief about the likelihood of each answer choice in isolation and the NLI model's beliefs about pair-wise answer choice compatibility. We show that a weighted MaxSAT solver can efficiently compute high-quality answer choices under this factor graph, improving over the raw model's predictions. Our experiments demonstrate that ConCoRD consistently boosts accuracy and consistency of off-the-shelf closed-book QA and VQA models using off-the-shelf NLI models, notably increasing accuracy of LXMERT on ConVQA by 5% absolute. See https://ericmitchell.ai/emnlp-2022-concord/ for code and data." @default.
- W4309869673 created "2022-11-29" @default.
- W4309869673 creator A5003324866 @default.
- W4309869673 creator A5005431772 @default.
- W4309869673 creator A5041328979 @default.
- W4309869673 creator A5046006076 @default.
- W4309869673 creator A5071017377 @default.
- W4309869673 creator A5074009646 @default.
- W4309869673 creator A5081688060 @default.
- W4309869673 creator A5091371424 @default.
- W4309869673 date "2022-11-21" @default.
- W4309869673 modified "2023-10-14" @default.
- W4309869673 title "Enhancing Self-Consistency and Performance of Pre-Trained Language Models through Natural Language Inference" @default.
- W4309869673 doi "https://doi.org/10.48550/arxiv.2211.11875" @default.
- W4309869673 hasPublicationYear "2022" @default.
- W4309869673 type Work @default.
- W4309869673 citedByCount "1" @default.
- W4309869673 countsByYear W43098696732023 @default.
- W4309869673 crossrefType "posted-content" @default.
- W4309869673 hasAuthorship W4309869673A5003324866 @default.
- W4309869673 hasAuthorship W4309869673A5005431772 @default.
- W4309869673 hasAuthorship W4309869673A5041328979 @default.
- W4309869673 hasAuthorship W4309869673A5046006076 @default.
- W4309869673 hasAuthorship W4309869673A5071017377 @default.
- W4309869673 hasAuthorship W4309869673A5074009646 @default.
- W4309869673 hasAuthorship W4309869673A5081688060 @default.
- W4309869673 hasAuthorship W4309869673A5091371424 @default.
- W4309869673 hasBestOaLocation W43098696731 @default.
- W4309869673 hasConcept C119857082 @default.
- W4309869673 hasConcept C137293760 @default.
- W4309869673 hasConcept C154945302 @default.
- W4309869673 hasConcept C204321447 @default.
- W4309869673 hasConcept C2776214188 @default.
- W4309869673 hasConcept C2776436953 @default.
- W4309869673 hasConcept C41008148 @default.
- W4309869673 hasConcept C44291984 @default.
- W4309869673 hasConceptScore W4309869673C119857082 @default.
- W4309869673 hasConceptScore W4309869673C137293760 @default.
- W4309869673 hasConceptScore W4309869673C154945302 @default.
- W4309869673 hasConceptScore W4309869673C204321447 @default.
- W4309869673 hasConceptScore W4309869673C2776214188 @default.
- W4309869673 hasConceptScore W4309869673C2776436953 @default.
- W4309869673 hasConceptScore W4309869673C41008148 @default.
- W4309869673 hasConceptScore W4309869673C44291984 @default.
- W4309869673 hasLocation W43098696731 @default.
- W4309869673 hasLocation W43098696732 @default.
- W4309869673 hasOpenAccess W4309869673 @default.
- W4309869673 hasPrimaryLocation W43098696731 @default.
- W4309869673 hasRelatedWork W1560657467 @default.
- W4309869673 hasRelatedWork W207304934 @default.
- W4309869673 hasRelatedWork W20999564 @default.
- W4309869673 hasRelatedWork W2359001871 @default.
- W4309869673 hasRelatedWork W3102273025 @default.
- W4309869673 hasRelatedWork W3207693618 @default.
- W4309869673 hasRelatedWork W4229058120 @default.
- W4309869673 hasRelatedWork W4297199749 @default.
- W4309869673 hasRelatedWork W4320486724 @default.
- W4309869673 hasRelatedWork W4377703168 @default.
- W4309869673 isParatext "false" @default.
- W4309869673 isRetracted "false" @default.
- W4309869673 workType "article" @default.