Matches in SemOpenAlex for { <https://semopenalex.org/work/W4287209352> ?p ?o ?g. }
Showing items 1 to 85 of
85
with 100 items per page.
- W4287209352 abstract "Large language models have shown promising results in zero-shot settings (Brown et al.,2020; Radford et al., 2019). For example, they can perform multiple choice tasks simply by conditioning on a question and selecting the answer with the highest probability. However, ranking by string probability can be problematic due to surface form competition-wherein different surface forms compete for probability mass, even if they represent the same underlying concept, e.g. computer and PC. Since probability mass is finite, this lowers the probability of the correct answer, due to competition from other strings that are valid answers (but not one of the multiple choice options). We introduce Domain Conditional Pointwise Mutual Information, an alternative scoring function that directly compensates for surface form competition by simply reweighing each option according to a term that is proportional to its a priori likelihood within the context of the specific zero-shot task. It achieves consistent gains in zero-shot performance over both calibrated (Zhao et al., 2021) and uncalibrated scoring functions on all GPT-2 and GPT-3 models over a variety of multiple choice datasets." @default.
- W4287209352 created "2022-07-25" @default.
- W4287209352 creator A5006531172 @default.
- W4287209352 creator A5045464993 @default.
- W4287209352 creator A5063151917 @default.
- W4287209352 creator A5067919401 @default.
- W4287209352 creator A5081420816 @default.
- W4287209352 date "2021-04-16" @default.
- W4287209352 modified "2023-09-28" @default.
- W4287209352 title "Surface Form Competition: Why the Highest Probability Answer Isn't Always Right" @default.
- W4287209352 doi "https://doi.org/10.48550/arxiv.2104.08315" @default.
- W4287209352 hasPublicationYear "2021" @default.
- W4287209352 type Work @default.
- W4287209352 citedByCount "0" @default.
- W4287209352 crossrefType "posted-content" @default.
- W4287209352 hasAuthorship W4287209352A5006531172 @default.
- W4287209352 hasAuthorship W4287209352A5045464993 @default.
- W4287209352 hasAuthorship W4287209352A5063151917 @default.
- W4287209352 hasAuthorship W4287209352A5067919401 @default.
- W4287209352 hasAuthorship W4287209352A5081420816 @default.
- W4287209352 hasBestOaLocation W42872093521 @default.
- W4287209352 hasConcept C105795698 @default.
- W4287209352 hasConcept C134306372 @default.
- W4287209352 hasConcept C138885662 @default.
- W4287209352 hasConcept C144237770 @default.
- W4287209352 hasConcept C149441793 @default.
- W4287209352 hasConcept C151730666 @default.
- W4287209352 hasConcept C154945302 @default.
- W4287209352 hasConcept C157486923 @default.
- W4287209352 hasConcept C162324750 @default.
- W4287209352 hasConcept C187736073 @default.
- W4287209352 hasConcept C18903297 @default.
- W4287209352 hasConcept C189430467 @default.
- W4287209352 hasConcept C197096303 @default.
- W4287209352 hasConcept C2777984123 @default.
- W4287209352 hasConcept C2779343474 @default.
- W4287209352 hasConcept C2780451532 @default.
- W4287209352 hasConcept C2780813799 @default.
- W4287209352 hasConcept C33923547 @default.
- W4287209352 hasConcept C37914503 @default.
- W4287209352 hasConcept C41008148 @default.
- W4287209352 hasConcept C41895202 @default.
- W4287209352 hasConcept C44492722 @default.
- W4287209352 hasConcept C86803240 @default.
- W4287209352 hasConcept C91306197 @default.
- W4287209352 hasConceptScore W4287209352C105795698 @default.
- W4287209352 hasConceptScore W4287209352C134306372 @default.
- W4287209352 hasConceptScore W4287209352C138885662 @default.
- W4287209352 hasConceptScore W4287209352C144237770 @default.
- W4287209352 hasConceptScore W4287209352C149441793 @default.
- W4287209352 hasConceptScore W4287209352C151730666 @default.
- W4287209352 hasConceptScore W4287209352C154945302 @default.
- W4287209352 hasConceptScore W4287209352C157486923 @default.
- W4287209352 hasConceptScore W4287209352C162324750 @default.
- W4287209352 hasConceptScore W4287209352C187736073 @default.
- W4287209352 hasConceptScore W4287209352C18903297 @default.
- W4287209352 hasConceptScore W4287209352C189430467 @default.
- W4287209352 hasConceptScore W4287209352C197096303 @default.
- W4287209352 hasConceptScore W4287209352C2777984123 @default.
- W4287209352 hasConceptScore W4287209352C2779343474 @default.
- W4287209352 hasConceptScore W4287209352C2780451532 @default.
- W4287209352 hasConceptScore W4287209352C2780813799 @default.
- W4287209352 hasConceptScore W4287209352C33923547 @default.
- W4287209352 hasConceptScore W4287209352C37914503 @default.
- W4287209352 hasConceptScore W4287209352C41008148 @default.
- W4287209352 hasConceptScore W4287209352C41895202 @default.
- W4287209352 hasConceptScore W4287209352C44492722 @default.
- W4287209352 hasConceptScore W4287209352C86803240 @default.
- W4287209352 hasConceptScore W4287209352C91306197 @default.
- W4287209352 hasLocation W42872093521 @default.
- W4287209352 hasOpenAccess W4287209352 @default.
- W4287209352 hasPrimaryLocation W42872093521 @default.
- W4287209352 hasRelatedWork W129790328 @default.
- W4287209352 hasRelatedWork W1523756637 @default.
- W4287209352 hasRelatedWork W193767348 @default.
- W4287209352 hasRelatedWork W2921932085 @default.
- W4287209352 hasRelatedWork W2975827855 @default.
- W4287209352 hasRelatedWork W3044665550 @default.
- W4287209352 hasRelatedWork W3099935545 @default.
- W4287209352 hasRelatedWork W4235623166 @default.
- W4287209352 hasRelatedWork W4288413872 @default.
- W4287209352 hasRelatedWork W60514588 @default.
- W4287209352 isParatext "false" @default.
- W4287209352 isRetracted "false" @default.
- W4287209352 workType "article" @default.