Matches in SemOpenAlex for { <https://semopenalex.org/work/W4206288797> ?p ?o ?g. }
Showing items 1 to 56 of
56
with 100 items per page.
- W4206288797 abstract "Existing techniques for mitigating dataset bias often leverage a biased model to identify biased instances. The role of these biased instances is then reduced during the training of the main model to enhance its robustness to out-of-distribution data. A common core assumption of these techniques is that the main model handles biased instances similarly to the biased model, in that it will resort to biases whenever available. In this paper, we show that this assumption does not hold in general. We carry out a critical investigation on two well-known datasets in the domain, MNLI and FEVER, along with two biased instance detection methods, partial-input and limited-capacity models. Our experiments show that in around a third to a half of instances, the biased model is unable to predict the main model’s behavior, highlighted by the significantly different parts of the input on which they base their decisions. Based on a manual validation, we also show that this estimate is highly in line with human interpretation. Our findings suggest that down-weighting of instances detected by bias detection methods, which is a widely-practiced procedure, is an unnecessary waste of training data. We release our code to facilitate reproducibility and future research." @default.
- W4206288797 created "2022-01-26" @default.
- W4206288797 creator A5024048254 @default.
- W4206288797 creator A5091017313 @default.
- W4206288797 date "2021-01-01" @default.
- W4206288797 modified "2023-10-18" @default.
- W4206288797 title "Don’t Discard All the Biased Instances: Investigating a Core Assumption in Dataset Bias Mitigation Techniques" @default.
- W4206288797 doi "https://doi.org/10.18653/v1/2021.findings-emnlp.405" @default.
- W4206288797 hasPublicationYear "2021" @default.
- W4206288797 type Work @default.
- W4206288797 citedByCount "0" @default.
- W4206288797 crossrefType "proceedings-article" @default.
- W4206288797 hasAuthorship W4206288797A5024048254 @default.
- W4206288797 hasAuthorship W4206288797A5091017313 @default.
- W4206288797 hasBestOaLocation W42062887971 @default.
- W4206288797 hasConcept C104317684 @default.
- W4206288797 hasConcept C119857082 @default.
- W4206288797 hasConcept C124101348 @default.
- W4206288797 hasConcept C126838900 @default.
- W4206288797 hasConcept C153083717 @default.
- W4206288797 hasConcept C154945302 @default.
- W4206288797 hasConcept C183115368 @default.
- W4206288797 hasConcept C185592680 @default.
- W4206288797 hasConcept C41008148 @default.
- W4206288797 hasConcept C55493867 @default.
- W4206288797 hasConcept C63479239 @default.
- W4206288797 hasConcept C71924100 @default.
- W4206288797 hasConceptScore W4206288797C104317684 @default.
- W4206288797 hasConceptScore W4206288797C119857082 @default.
- W4206288797 hasConceptScore W4206288797C124101348 @default.
- W4206288797 hasConceptScore W4206288797C126838900 @default.
- W4206288797 hasConceptScore W4206288797C153083717 @default.
- W4206288797 hasConceptScore W4206288797C154945302 @default.
- W4206288797 hasConceptScore W4206288797C183115368 @default.
- W4206288797 hasConceptScore W4206288797C185592680 @default.
- W4206288797 hasConceptScore W4206288797C41008148 @default.
- W4206288797 hasConceptScore W4206288797C55493867 @default.
- W4206288797 hasConceptScore W4206288797C63479239 @default.
- W4206288797 hasConceptScore W4206288797C71924100 @default.
- W4206288797 hasLocation W42062887971 @default.
- W4206288797 hasLocation W42062887972 @default.
- W4206288797 hasOpenAccess W4206288797 @default.
- W4206288797 hasPrimaryLocation W42062887971 @default.
- W4206288797 hasRelatedWork W1556035623 @default.
- W4206288797 hasRelatedWork W1999357716 @default.
- W4206288797 hasRelatedWork W2066966161 @default.
- W4206288797 hasRelatedWork W2369301223 @default.
- W4206288797 hasRelatedWork W2584455473 @default.
- W4206288797 hasRelatedWork W2961085424 @default.
- W4206288797 hasRelatedWork W4286629047 @default.
- W4206288797 hasRelatedWork W4306321456 @default.
- W4206288797 hasRelatedWork W4306674287 @default.
- W4206288797 hasRelatedWork W4224009465 @default.
- W4206288797 isParatext "false" @default.
- W4206288797 isRetracted "false" @default.
- W4206288797 workType "article" @default.