Matches in SemOpenAlex for { <https://semopenalex.org/work/W4378465178> ?p ?o ?g. }
Showing items 1 to 59 of
59
with 100 items per page.
- W4378465178 abstract "Editing model parameters directly in Transformers makes updating black-box models possible without re-training (Meng et al., 2023). However, these editing methods have only been evaluated on statements about encyclopedic knowledge with a single correct answer. Commonsense knowledge with multiple correct answers, e.g., an apple can be green or red but not transparent, has not been studied but is as essential for enhancing transformers' reliability and usefulness. In this paper, we investigate whether commonsense judgments are causally associated with localized, editable parameters in Transformers, and we provide an affirmative answer. We find that directly applying the MEMIT editing algorithm results in sub-par performance and improve it for the commonsense domain by varying edit tokens and improving the layer selection strategy, i.e., $MEMIT_{CSK}$. GPT-2 Large and XL models edited using $MEMIT_{CSK}$ outperform best-fine-tuned baselines by 10.97% and 10.73% F1 scores on PEP3k and 20Q datasets. In addition, we propose a novel evaluation dataset, PROBE SET, that contains unaffected and affected neighborhoods, affected paraphrases, and affected reasoning challenges. $MEMIT_{CSK}$ performs well across the metrics while fine-tuning baselines show significant trade-offs between unaffected and affected metrics. These results suggest a compelling future direction for incorporating feedback about common sense into Transformers through direct model editing." @default.
- W4378465178 created "2023-05-27" @default.
- W4378465178 creator A5016133514 @default.
- W4378465178 creator A5022684697 @default.
- W4378465178 creator A5033982504 @default.
- W4378465178 creator A5034858125 @default.
- W4378465178 creator A5047755233 @default.
- W4378465178 creator A5052882677 @default.
- W4378465178 creator A5079108871 @default.
- W4378465178 date "2023-05-24" @default.
- W4378465178 modified "2023-10-13" @default.
- W4378465178 title "Editing Common Sense in Transformers" @default.
- W4378465178 doi "https://doi.org/10.48550/arxiv.2305.14956" @default.
- W4378465178 hasPublicationYear "2023" @default.
- W4378465178 type Work @default.
- W4378465178 citedByCount "0" @default.
- W4378465178 crossrefType "posted-content" @default.
- W4378465178 hasAuthorship W4378465178A5016133514 @default.
- W4378465178 hasAuthorship W4378465178A5022684697 @default.
- W4378465178 hasAuthorship W4378465178A5033982504 @default.
- W4378465178 hasAuthorship W4378465178A5034858125 @default.
- W4378465178 hasAuthorship W4378465178A5047755233 @default.
- W4378465178 hasAuthorship W4378465178A5052882677 @default.
- W4378465178 hasAuthorship W4378465178A5079108871 @default.
- W4378465178 hasBestOaLocation W43784651781 @default.
- W4378465178 hasConcept C119599485 @default.
- W4378465178 hasConcept C119857082 @default.
- W4378465178 hasConcept C127413603 @default.
- W4378465178 hasConcept C154945302 @default.
- W4378465178 hasConcept C165801399 @default.
- W4378465178 hasConcept C193221554 @default.
- W4378465178 hasConcept C204321447 @default.
- W4378465178 hasConcept C41008148 @default.
- W4378465178 hasConcept C66322947 @default.
- W4378465178 hasConceptScore W4378465178C119599485 @default.
- W4378465178 hasConceptScore W4378465178C119857082 @default.
- W4378465178 hasConceptScore W4378465178C127413603 @default.
- W4378465178 hasConceptScore W4378465178C154945302 @default.
- W4378465178 hasConceptScore W4378465178C165801399 @default.
- W4378465178 hasConceptScore W4378465178C193221554 @default.
- W4378465178 hasConceptScore W4378465178C204321447 @default.
- W4378465178 hasConceptScore W4378465178C41008148 @default.
- W4378465178 hasConceptScore W4378465178C66322947 @default.
- W4378465178 hasLocation W43784651781 @default.
- W4378465178 hasOpenAccess W4378465178 @default.
- W4378465178 hasPrimaryLocation W43784651781 @default.
- W4378465178 hasRelatedWork W2961085424 @default.
- W4378465178 hasRelatedWork W3046775127 @default.
- W4378465178 hasRelatedWork W3107602296 @default.
- W4378465178 hasRelatedWork W3170094116 @default.
- W4378465178 hasRelatedWork W3209574120 @default.
- W4378465178 hasRelatedWork W4210805261 @default.
- W4378465178 hasRelatedWork W4306674287 @default.
- W4378465178 hasRelatedWork W4312192474 @default.
- W4378465178 hasRelatedWork W4386462264 @default.
- W4378465178 hasRelatedWork W4387297750 @default.
- W4378465178 isParatext "false" @default.
- W4378465178 isRetracted "false" @default.
- W4378465178 workType "article" @default.