Matches in SemOpenAlex for { <https://semopenalex.org/work/W3201957104> ?p ?o ?g. }
- W3201957104 abstract "We introduce an evaluation methodology for visual question answering (VQA) to better diagnose cases of shortcut learning. These cases happen when a model exploits spurious statistical regularities to produce correct answers but does not actually deploy the desired behavior. There is a need to identify possible shortcuts in a dataset and assess their use before deploying a model in the real world. The research community in VQA has focused exclusively on question-based shortcuts, where a model might, for example, answer What is the color of the sky with blue by relying mostly on the question-conditional training prior and give little weight to visual evidence. We go a step further and consider multimodal shortcuts that involve both questions and images. We first identify potential shortcuts in the popular VQA v2 training set by mining trivial predictive rules such as co-occurrences of words and visual elements. We then introduce VQA-CounterExamples (VQACE), an evaluation protocol based on our subset of CounterExamples i.e. image-question-answer triplets where our rules lead to incorrect answers. We use this new evaluation in a large-scale study of existing approaches for VQA. We demonstrate that even state-of-the-art models perform poorly and that existing techniques to reduce biases are largely ineffective in this context. Our findings suggest that past work on question-based biases in VQA has only addressed one facet of a complex issue. The code for our method is available at https://github.com/cdancette/detect-shortcuts" @default.
- W3201957104 created "2021-10-11" @default.
- W3201957104 creator A5022871131 @default.
- W3201957104 creator A5036295862 @default.
- W3201957104 creator A5046767053 @default.
- W3201957104 creator A5067549788 @default.
- W3201957104 date "2021-10-01" @default.
- W3201957104 modified "2023-10-01" @default.
- W3201957104 title "Beyond Question-Based Biases: Assessing Multimodal Shortcut Learning in Visual Question Answering" @default.
- W3201957104 cites W1933349210 @default.
- W3201957104 cites W2282821441 @default.
- W3201957104 cites W2560730294 @default.
- W3201957104 cites W2561715562 @default.
- W3201957104 cites W2597425697 @default.
- W3201957104 cites W2745461083 @default.
- W3201957104 cites W2789240707 @default.
- W3201957104 cites W2886641317 @default.
- W3201957104 cites W2895472239 @default.
- W3201957104 cites W2940753425 @default.
- W3201957104 cites W2951286828 @default.
- W3201957104 cites W2962884579 @default.
- W3201957104 cites W2963224792 @default.
- W3201957104 cites W2963518342 @default.
- W3201957104 cites W2963717374 @default.
- W3201957104 cites W2963954913 @default.
- W3201957104 cites W2963969878 @default.
- W3201957104 cites W2964072591 @default.
- W3201957104 cites W2964118342 @default.
- W3201957104 cites W2965628639 @default.
- W3201957104 cites W2968993450 @default.
- W3201957104 cites W2982699810 @default.
- W3201957104 cites W3016970897 @default.
- W3201957104 cites W3035517717 @default.
- W3201957104 cites W3035561630 @default.
- W3201957104 cites W3101609372 @default.
- W3201957104 cites W3104788521 @default.
- W3201957104 doi "https://doi.org/10.1109/iccv48922.2021.00160" @default.
- W3201957104 hasPublicationYear "2021" @default.
- W3201957104 type Work @default.
- W3201957104 sameAs 3201957104 @default.
- W3201957104 citedByCount "14" @default.
- W3201957104 countsByYear W32019571042022 @default.
- W3201957104 countsByYear W32019571042023 @default.
- W3201957104 crossrefType "proceedings-article" @default.
- W3201957104 hasAuthorship W3201957104A5022871131 @default.
- W3201957104 hasAuthorship W3201957104A5036295862 @default.
- W3201957104 hasAuthorship W3201957104A5046767053 @default.
- W3201957104 hasAuthorship W3201957104A5067549788 @default.
- W3201957104 hasBestOaLocation W32019571044 @default.
- W3201957104 hasConcept C111919701 @default.
- W3201957104 hasConcept C118615104 @default.
- W3201957104 hasConcept C119857082 @default.
- W3201957104 hasConcept C142724271 @default.
- W3201957104 hasConcept C151730666 @default.
- W3201957104 hasConcept C154945302 @default.
- W3201957104 hasConcept C162838799 @default.
- W3201957104 hasConcept C165696696 @default.
- W3201957104 hasConcept C177264268 @default.
- W3201957104 hasConcept C199360897 @default.
- W3201957104 hasConcept C204787440 @default.
- W3201957104 hasConcept C23123220 @default.
- W3201957104 hasConcept C2776760102 @default.
- W3201957104 hasConcept C2779343474 @default.
- W3201957104 hasConcept C2780385302 @default.
- W3201957104 hasConcept C33923547 @default.
- W3201957104 hasConcept C38652104 @default.
- W3201957104 hasConcept C41008148 @default.
- W3201957104 hasConcept C43126263 @default.
- W3201957104 hasConcept C44291984 @default.
- W3201957104 hasConcept C51929080 @default.
- W3201957104 hasConcept C71924100 @default.
- W3201957104 hasConcept C86803240 @default.
- W3201957104 hasConcept C97256817 @default.
- W3201957104 hasConceptScore W3201957104C111919701 @default.
- W3201957104 hasConceptScore W3201957104C118615104 @default.
- W3201957104 hasConceptScore W3201957104C119857082 @default.
- W3201957104 hasConceptScore W3201957104C142724271 @default.
- W3201957104 hasConceptScore W3201957104C151730666 @default.
- W3201957104 hasConceptScore W3201957104C154945302 @default.
- W3201957104 hasConceptScore W3201957104C162838799 @default.
- W3201957104 hasConceptScore W3201957104C165696696 @default.
- W3201957104 hasConceptScore W3201957104C177264268 @default.
- W3201957104 hasConceptScore W3201957104C199360897 @default.
- W3201957104 hasConceptScore W3201957104C204787440 @default.
- W3201957104 hasConceptScore W3201957104C23123220 @default.
- W3201957104 hasConceptScore W3201957104C2776760102 @default.
- W3201957104 hasConceptScore W3201957104C2779343474 @default.
- W3201957104 hasConceptScore W3201957104C2780385302 @default.
- W3201957104 hasConceptScore W3201957104C33923547 @default.
- W3201957104 hasConceptScore W3201957104C38652104 @default.
- W3201957104 hasConceptScore W3201957104C41008148 @default.
- W3201957104 hasConceptScore W3201957104C43126263 @default.
- W3201957104 hasConceptScore W3201957104C44291984 @default.
- W3201957104 hasConceptScore W3201957104C51929080 @default.
- W3201957104 hasConceptScore W3201957104C71924100 @default.
- W3201957104 hasConceptScore W3201957104C86803240 @default.
- W3201957104 hasConceptScore W3201957104C97256817 @default.
- W3201957104 hasLocation W32019571041 @default.
- W3201957104 hasLocation W32019571042 @default.
- W3201957104 hasLocation W32019571043 @default.