Matches in SemOpenAlex for { <https://semopenalex.org/work/W4385572016> ?p ?o ?g. }
Showing items 1 to 67 of
67
with 100 items per page.
- W4385572016 abstract "Large language models (LMs) beyond a certain scale, demonstrate the emergent capability of generating free-text rationales for their predictions via chain-of-thought (CoT) prompting.While CoT can yield dramatically improved performance, such gains are only observed for sufficiently large LMs. Even more concerning, there is little guarantee that the generated rationales are consistent with LM’s predictions or faithfully justify the decisions. In this work, we propose SCOTT, a faithful knowledge distillation method to learn a small, self-consistent CoT model from a teacher model that is orders of magnitude larger. To form better supervision, we elicit rationales supporting the gold answers from a large LM (teacher) by contrastive decoding, which encourages the teacher to generate tokens that become more plausible only when the answer is considered. To ensure faithful distillation, we use the teacher-generated rationales to learn a student LM with a counterfactual reasoning objective, which prevents the student from ignoring the rationales to make inconsistent predictions. Experiments show that while yielding comparable performance, our method leads to a more faithful model than baselines. Further analysis shows that such a model respects the rationales more when making decisions; thus, we can improve its performance more by refining its rationales." @default.
- W4385572016 created "2023-08-05" @default.
- W4385572016 creator A5009408707 @default.
- W4385572016 creator A5028075867 @default.
- W4385572016 creator A5029762051 @default.
- W4385572016 creator A5045927598 @default.
- W4385572016 creator A5060335470 @default.
- W4385572016 creator A5088485347 @default.
- W4385572016 date "2023-01-01" @default.
- W4385572016 modified "2023-10-12" @default.
- W4385572016 title "SCOTT: Self-Consistent Chain-of-Thought Distillation" @default.
- W4385572016 doi "https://doi.org/10.18653/v1/2023.acl-long.304" @default.
- W4385572016 hasPublicationYear "2023" @default.
- W4385572016 type Work @default.
- W4385572016 citedByCount "1" @default.
- W4385572016 crossrefType "proceedings-article" @default.
- W4385572016 hasAuthorship W4385572016A5009408707 @default.
- W4385572016 hasAuthorship W4385572016A5028075867 @default.
- W4385572016 hasAuthorship W4385572016A5029762051 @default.
- W4385572016 hasAuthorship W4385572016A5045927598 @default.
- W4385572016 hasAuthorship W4385572016A5060335470 @default.
- W4385572016 hasAuthorship W4385572016A5088485347 @default.
- W4385572016 hasBestOaLocation W43855720161 @default.
- W4385572016 hasConcept C105795698 @default.
- W4385572016 hasConcept C108650721 @default.
- W4385572016 hasConcept C121332964 @default.
- W4385572016 hasConcept C154945302 @default.
- W4385572016 hasConcept C15744967 @default.
- W4385572016 hasConcept C165064840 @default.
- W4385572016 hasConcept C178790620 @default.
- W4385572016 hasConcept C185592680 @default.
- W4385572016 hasConcept C204030448 @default.
- W4385572016 hasConcept C2778755073 @default.
- W4385572016 hasConcept C33923547 @default.
- W4385572016 hasConcept C41008148 @default.
- W4385572016 hasConcept C62520636 @default.
- W4385572016 hasConcept C77805123 @default.
- W4385572016 hasConceptScore W4385572016C105795698 @default.
- W4385572016 hasConceptScore W4385572016C108650721 @default.
- W4385572016 hasConceptScore W4385572016C121332964 @default.
- W4385572016 hasConceptScore W4385572016C154945302 @default.
- W4385572016 hasConceptScore W4385572016C15744967 @default.
- W4385572016 hasConceptScore W4385572016C165064840 @default.
- W4385572016 hasConceptScore W4385572016C178790620 @default.
- W4385572016 hasConceptScore W4385572016C185592680 @default.
- W4385572016 hasConceptScore W4385572016C204030448 @default.
- W4385572016 hasConceptScore W4385572016C2778755073 @default.
- W4385572016 hasConceptScore W4385572016C33923547 @default.
- W4385572016 hasConceptScore W4385572016C41008148 @default.
- W4385572016 hasConceptScore W4385572016C62520636 @default.
- W4385572016 hasConceptScore W4385572016C77805123 @default.
- W4385572016 hasLocation W43855720161 @default.
- W4385572016 hasOpenAccess W4385572016 @default.
- W4385572016 hasPrimaryLocation W43855720161 @default.
- W4385572016 hasRelatedWork W1805447287 @default.
- W4385572016 hasRelatedWork W2300845814 @default.
- W4385572016 hasRelatedWork W3122186339 @default.
- W4385572016 hasRelatedWork W3124566012 @default.
- W4385572016 hasRelatedWork W3147935347 @default.
- W4385572016 hasRelatedWork W4225851326 @default.
- W4385572016 hasRelatedWork W4282029540 @default.
- W4385572016 hasRelatedWork W4292258433 @default.
- W4385572016 hasRelatedWork W4292957081 @default.
- W4385572016 hasRelatedWork W4361769030 @default.
- W4385572016 isParatext "false" @default.
- W4385572016 isRetracted "false" @default.
- W4385572016 workType "article" @default.