Matches in SemOpenAlex for { <https://semopenalex.org/work/W4297777863> ?p ?o ?g. }
Showing items 1 to 81 of
81
with 100 items per page.
- W4297777863 abstract "An extractive rationale explains a language model's (LM's) prediction on a given task instance by highlighting the text inputs that most influenced the prediction. Ideally, rationale extraction should be faithful (reflective of LM's actual behavior) and plausible (convincing to humans), without compromising the LM's (i.e., task model's) task performance. Although attribution algorithms and select-predict pipelines are commonly used in rationale extraction, they both rely on certain heuristics that hinder them from satisfying all three desiderata. In light of this, we propose UNIREX, a flexible learning framework that generalizes rationale extractor optimization as follows: (1) specify architecture for a learned rationale extractor; (2) select explainability objectives (i.e., faithfulness and plausibility criteria); and (3) jointly the train task model and rationale extractor on the task using the selected objectives. UNIREX enables replacing prior works' heuristic design choices with a generic learned rationale extractor in (1) and optimizing it for all three desiderata in (2)-(3). To facilitate comparison between methods with respect to multiple desiderata, we introduce the Normalized Relative Gain (NRG) metric. Across five text classification datasets, our best UNIREX configuration outperforms baselines by an average of 32.9% NRG. Plus, we find that UNIREX-trained rationale extractors can even generalize to unseen datasets and tasks." @default.
- W4297777863 created "2022-10-01" @default.
- W4297777863 creator A5009408707 @default.
- W4297777863 creator A5018913504 @default.
- W4297777863 creator A5030713034 @default.
- W4297777863 creator A5041564394 @default.
- W4297777863 creator A5043413179 @default.
- W4297777863 creator A5046195860 @default.
- W4297777863 creator A5051519610 @default.
- W4297777863 creator A5067685509 @default.
- W4297777863 date "2021-12-16" @default.
- W4297777863 modified "2023-10-01" @default.
- W4297777863 title "UNIREX: A Unified Learning Framework for Language Model Rationale Extraction" @default.
- W4297777863 doi "https://doi.org/10.48550/arxiv.2112.08802" @default.
- W4297777863 hasPublicationYear "2021" @default.
- W4297777863 type Work @default.
- W4297777863 citedByCount "0" @default.
- W4297777863 crossrefType "posted-content" @default.
- W4297777863 hasAuthorship W4297777863A5009408707 @default.
- W4297777863 hasAuthorship W4297777863A5018913504 @default.
- W4297777863 hasAuthorship W4297777863A5030713034 @default.
- W4297777863 hasAuthorship W4297777863A5041564394 @default.
- W4297777863 hasAuthorship W4297777863A5043413179 @default.
- W4297777863 hasAuthorship W4297777863A5046195860 @default.
- W4297777863 hasAuthorship W4297777863A5051519610 @default.
- W4297777863 hasAuthorship W4297777863A5067685509 @default.
- W4297777863 hasBestOaLocation W42977778631 @default.
- W4297777863 hasConcept C105795698 @default.
- W4297777863 hasConcept C111919701 @default.
- W4297777863 hasConcept C117978034 @default.
- W4297777863 hasConcept C119857082 @default.
- W4297777863 hasConcept C127413603 @default.
- W4297777863 hasConcept C127705205 @default.
- W4297777863 hasConcept C137293760 @default.
- W4297777863 hasConcept C154945302 @default.
- W4297777863 hasConcept C162324750 @default.
- W4297777863 hasConcept C165064840 @default.
- W4297777863 hasConcept C173801870 @default.
- W4297777863 hasConcept C176217482 @default.
- W4297777863 hasConcept C187736073 @default.
- W4297777863 hasConcept C204321447 @default.
- W4297777863 hasConcept C21547014 @default.
- W4297777863 hasConcept C21880701 @default.
- W4297777863 hasConcept C2780451532 @default.
- W4297777863 hasConcept C33923547 @default.
- W4297777863 hasConcept C41008148 @default.
- W4297777863 hasConceptScore W4297777863C105795698 @default.
- W4297777863 hasConceptScore W4297777863C111919701 @default.
- W4297777863 hasConceptScore W4297777863C117978034 @default.
- W4297777863 hasConceptScore W4297777863C119857082 @default.
- W4297777863 hasConceptScore W4297777863C127413603 @default.
- W4297777863 hasConceptScore W4297777863C127705205 @default.
- W4297777863 hasConceptScore W4297777863C137293760 @default.
- W4297777863 hasConceptScore W4297777863C154945302 @default.
- W4297777863 hasConceptScore W4297777863C162324750 @default.
- W4297777863 hasConceptScore W4297777863C165064840 @default.
- W4297777863 hasConceptScore W4297777863C173801870 @default.
- W4297777863 hasConceptScore W4297777863C176217482 @default.
- W4297777863 hasConceptScore W4297777863C187736073 @default.
- W4297777863 hasConceptScore W4297777863C204321447 @default.
- W4297777863 hasConceptScore W4297777863C21547014 @default.
- W4297777863 hasConceptScore W4297777863C21880701 @default.
- W4297777863 hasConceptScore W4297777863C2780451532 @default.
- W4297777863 hasConceptScore W4297777863C33923547 @default.
- W4297777863 hasConceptScore W4297777863C41008148 @default.
- W4297777863 hasLocation W42977778631 @default.
- W4297777863 hasOpenAccess W4297777863 @default.
- W4297777863 hasPrimaryLocation W42977778631 @default.
- W4297777863 hasRelatedWork W1068348394 @default.
- W4297777863 hasRelatedWork W1766024167 @default.
- W4297777863 hasRelatedWork W2009872548 @default.
- W4297777863 hasRelatedWork W2081647779 @default.
- W4297777863 hasRelatedWork W2086599114 @default.
- W4297777863 hasRelatedWork W2965145359 @default.
- W4297777863 hasRelatedWork W3107474891 @default.
- W4297777863 hasRelatedWork W3125601490 @default.
- W4297777863 hasRelatedWork W3185852197 @default.
- W4297777863 hasRelatedWork W3202215796 @default.
- W4297777863 isParatext "false" @default.
- W4297777863 isRetracted "false" @default.
- W4297777863 workType "article" @default.