Matches in SemOpenAlex for { <https://semopenalex.org/work/W3136788995> ?p ?o ?g. }
- W3136788995 abstract "Recent works have shown that supervised models often exploit data artifacts to achieve good test scores while their performance severely degrades on samples outside their training distribution. Contrast sets (Gardneret al., 2020) quantify this phenomenon by perturbing test samples in a minimal way such that the output label is modified. While most contrast sets were created manually, requiring intensive annotation effort, we present a novel method which leverages rich semantic input representation to automatically generate contrast sets for the visual question answering task. Our method computes the answer of perturbed questions, thus vastly reducing annotation cost and enabling thorough evaluation of models' performance on various semantic aspects (e.g., spatial or relational reasoning). We demonstrate the effectiveness of our approach on the GQA dataset and its semantic scene graph image representation. We find that, despite GQA's compositionality and carefully balanced label distribution, two high-performing models drop 13-17% in accuracy compared to the original test set. Finally, we show that our automatic perturbation can be applied to the training set to mitigate the degradation in performance, opening the door to more robust models." @default.
- W3136788995 created "2021-03-29" @default.
- W3136788995 creator A5007903277 @default.
- W3136788995 creator A5013658527 @default.
- W3136788995 creator A5068580969 @default.
- W3136788995 creator A5082136238 @default.
- W3136788995 date "2021-03-17" @default.
- W3136788995 modified "2023-09-28" @default.
- W3136788995 title "Automatic Generation of Contrast Sets from Scene Graphs: Probing the Compositional Consistency of GQA" @default.
- W3136788995 cites W2277195237 @default.
- W3136788995 cites W2560730294 @default.
- W3136788995 cites W2786209943 @default.
- W3136788995 cites W2799007037 @default.
- W3136788995 cites W2946609015 @default.
- W3136788995 cites W2950104027 @default.
- W3136788995 cites W2950470622 @default.
- W3136788995 cites W2951286828 @default.
- W3136788995 cites W2952328691 @default.
- W3136788995 cites W2962749469 @default.
- W3136788995 cites W2963159690 @default.
- W3136788995 cites W2963890019 @default.
- W3136788995 cites W2970231061 @default.
- W3136788995 cites W2970442950 @default.
- W3136788995 cites W2977235550 @default.
- W3136788995 cites W3039185146 @default.
- W3136788995 cites W3093188800 @default.
- W3136788995 cites W3101902264 @default.
- W3136788995 cites W3105928338 @default.
- W3136788995 hasPublicationYear "2021" @default.
- W3136788995 type Work @default.
- W3136788995 sameAs 3136788995 @default.
- W3136788995 citedByCount "0" @default.
- W3136788995 crossrefType "posted-content" @default.
- W3136788995 hasAuthorship W3136788995A5007903277 @default.
- W3136788995 hasAuthorship W3136788995A5013658527 @default.
- W3136788995 hasAuthorship W3136788995A5068580969 @default.
- W3136788995 hasAuthorship W3136788995A5082136238 @default.
- W3136788995 hasConcept C119857082 @default.
- W3136788995 hasConcept C121375916 @default.
- W3136788995 hasConcept C124101348 @default.
- W3136788995 hasConcept C132525143 @default.
- W3136788995 hasConcept C153180895 @default.
- W3136788995 hasConcept C154945302 @default.
- W3136788995 hasConcept C165696696 @default.
- W3136788995 hasConcept C169903167 @default.
- W3136788995 hasConcept C177264268 @default.
- W3136788995 hasConcept C17744445 @default.
- W3136788995 hasConcept C179372163 @default.
- W3136788995 hasConcept C199360897 @default.
- W3136788995 hasConcept C199539241 @default.
- W3136788995 hasConcept C204321447 @default.
- W3136788995 hasConcept C205711294 @default.
- W3136788995 hasConcept C2776321320 @default.
- W3136788995 hasConcept C2776359362 @default.
- W3136788995 hasConcept C2776436953 @default.
- W3136788995 hasConcept C2776502983 @default.
- W3136788995 hasConcept C38652104 @default.
- W3136788995 hasConcept C41008148 @default.
- W3136788995 hasConcept C51632099 @default.
- W3136788995 hasConcept C80444323 @default.
- W3136788995 hasConcept C94625758 @default.
- W3136788995 hasConceptScore W3136788995C119857082 @default.
- W3136788995 hasConceptScore W3136788995C121375916 @default.
- W3136788995 hasConceptScore W3136788995C124101348 @default.
- W3136788995 hasConceptScore W3136788995C132525143 @default.
- W3136788995 hasConceptScore W3136788995C153180895 @default.
- W3136788995 hasConceptScore W3136788995C154945302 @default.
- W3136788995 hasConceptScore W3136788995C165696696 @default.
- W3136788995 hasConceptScore W3136788995C169903167 @default.
- W3136788995 hasConceptScore W3136788995C177264268 @default.
- W3136788995 hasConceptScore W3136788995C17744445 @default.
- W3136788995 hasConceptScore W3136788995C179372163 @default.
- W3136788995 hasConceptScore W3136788995C199360897 @default.
- W3136788995 hasConceptScore W3136788995C199539241 @default.
- W3136788995 hasConceptScore W3136788995C204321447 @default.
- W3136788995 hasConceptScore W3136788995C205711294 @default.
- W3136788995 hasConceptScore W3136788995C2776321320 @default.
- W3136788995 hasConceptScore W3136788995C2776359362 @default.
- W3136788995 hasConceptScore W3136788995C2776436953 @default.
- W3136788995 hasConceptScore W3136788995C2776502983 @default.
- W3136788995 hasConceptScore W3136788995C38652104 @default.
- W3136788995 hasConceptScore W3136788995C41008148 @default.
- W3136788995 hasConceptScore W3136788995C51632099 @default.
- W3136788995 hasConceptScore W3136788995C80444323 @default.
- W3136788995 hasConceptScore W3136788995C94625758 @default.
- W3136788995 hasLocation W31367889951 @default.
- W3136788995 hasOpenAccess W3136788995 @default.
- W3136788995 hasPrimaryLocation W31367889951 @default.
- W3136788995 hasRelatedWork W2137123920 @default.
- W3136788995 hasRelatedWork W2759729257 @default.
- W3136788995 hasRelatedWork W2773708732 @default.
- W3136788995 hasRelatedWork W2782368579 @default.
- W3136788995 hasRelatedWork W2889435835 @default.
- W3136788995 hasRelatedWork W2896107389 @default.
- W3136788995 hasRelatedWork W2903619461 @default.
- W3136788995 hasRelatedWork W2918598763 @default.
- W3136788995 hasRelatedWork W2941109201 @default.
- W3136788995 hasRelatedWork W2948636084 @default.
- W3136788995 hasRelatedWork W2950483904 @default.
- W3136788995 hasRelatedWork W2992195701 @default.