Matches in SemOpenAlex for { <https://semopenalex.org/work/W2981202322> ?p ?o ?g. }
- W2981202322 abstract "An increasing amount of studies have investigated the decision-making process of VQA models. Many of these studies focus on the reason behind the correct answer chosen by a model. Yet, the reason why the distracting answer chose by a model has rarely been studied. To this end, we introduce a novel task called textit{textual Distractors Generation for VQA} (DG-VQA) that explaining the decision boundaries of existing VQA models. The goal of DG-VQA is to generate the most confusing set of textual distractors in multi-choice VQA tasks which expose the vulnerability of existing models (i.e. to generate distractors that lure existing models to fail). We show that DG-VQA can be formulated as a Markov Decision Process, and present a reinforcement learning solution to come up with distractors in an unsupervised manner. The solution addresses the lack of large annotated corpus issues in previous distractor generation methods. Our proposed model receives reward signals from fully-trained multi-choice VQA models and updates its parameters via policy gradient. The empirical results show that the generated textual distractors can successfully attack several popular VQA models with an average $20%$ accuracy drop from $64%$. Furthermore, we conduct adversarial training to improve the robustness of VQA models by incorporating the generated distractors. Empirical results validate the effectiveness of adversarial training by showing a performance improvement of $27%$ for the multi-choice VQA task." @default.
- W2981202322 created "2019-10-25" @default.
- W2981202322 creator A5002278578 @default.
- W2981202322 creator A5019288987 @default.
- W2981202322 creator A5036815832 @default.
- W2981202322 creator A5060501596 @default.
- W2981202322 date "2019-10-21" @default.
- W2981202322 modified "2023-09-29" @default.
- W2981202322 title "Good, Better, Best: Textual Distractors Generation for Multi-Choice VQA via Policy Gradient." @default.
- W2981202322 cites W1933349210 @default.
- W2981202322 cites W1983927101 @default.
- W2981202322 cites W2099471712 @default.
- W2981202322 cites W2119717200 @default.
- W2981202322 cites W2121863487 @default.
- W2981202322 cites W2129496160 @default.
- W2981202322 cites W2145482038 @default.
- W2981202322 cites W2153579005 @default.
- W2981202322 cites W2194775991 @default.
- W2981202322 cites W2243397390 @default.
- W2981202322 cites W2251056936 @default.
- W2981202322 cites W2251601342 @default.
- W2981202322 cites W2405756170 @default.
- W2981202322 cites W2471094925 @default.
- W2981202322 cites W2493916176 @default.
- W2981202322 cites W2758592353 @default.
- W2981202322 cites W2770171824 @default.
- W2981202322 cites W2772621923 @default.
- W2981202322 cites W2792263949 @default.
- W2981202322 cites W2805070499 @default.
- W2981202322 cites W2809669277 @default.
- W2981202322 cites W2825507547 @default.
- W2981202322 cites W2887978863 @default.
- W2981202322 cites W2902900716 @default.
- W2981202322 cites W2912087091 @default.
- W2981202322 cites W2919115771 @default.
- W2981202322 cites W2952085755 @default.
- W2981202322 cites W2962718684 @default.
- W2981202322 cites W2962749469 @default.
- W2981202322 cites W2962835968 @default.
- W2981202322 cites W2963051286 @default.
- W2981202322 cites W2963084599 @default.
- W2981202322 cites W2963115613 @default.
- W2981202322 cites W2963207607 @default.
- W2981202322 cites W2963277051 @default.
- W2981202322 cites W2963341956 @default.
- W2981202322 cites W2963352593 @default.
- W2981202322 cites W2963383024 @default.
- W2981202322 cites W2964087080 @default.
- W2981202322 cites W2964153729 @default.
- W2981202322 cites W3110388292 @default.
- W2981202322 hasPublicationYear "2019" @default.
- W2981202322 type Work @default.
- W2981202322 sameAs 2981202322 @default.
- W2981202322 citedByCount "0" @default.
- W2981202322 crossrefType "posted-content" @default.
- W2981202322 hasAuthorship W2981202322A5002278578 @default.
- W2981202322 hasAuthorship W2981202322A5019288987 @default.
- W2981202322 hasAuthorship W2981202322A5036815832 @default.
- W2981202322 hasAuthorship W2981202322A5060501596 @default.
- W2981202322 hasConcept C104317684 @default.
- W2981202322 hasConcept C111919701 @default.
- W2981202322 hasConcept C119857082 @default.
- W2981202322 hasConcept C154945302 @default.
- W2981202322 hasConcept C162324750 @default.
- W2981202322 hasConcept C177264268 @default.
- W2981202322 hasConcept C185592680 @default.
- W2981202322 hasConcept C187736073 @default.
- W2981202322 hasConcept C199360897 @default.
- W2981202322 hasConcept C2780451532 @default.
- W2981202322 hasConcept C37736160 @default.
- W2981202322 hasConcept C41008148 @default.
- W2981202322 hasConcept C55493867 @default.
- W2981202322 hasConcept C63479239 @default.
- W2981202322 hasConcept C98045186 @default.
- W2981202322 hasConceptScore W2981202322C104317684 @default.
- W2981202322 hasConceptScore W2981202322C111919701 @default.
- W2981202322 hasConceptScore W2981202322C119857082 @default.
- W2981202322 hasConceptScore W2981202322C154945302 @default.
- W2981202322 hasConceptScore W2981202322C162324750 @default.
- W2981202322 hasConceptScore W2981202322C177264268 @default.
- W2981202322 hasConceptScore W2981202322C185592680 @default.
- W2981202322 hasConceptScore W2981202322C187736073 @default.
- W2981202322 hasConceptScore W2981202322C199360897 @default.
- W2981202322 hasConceptScore W2981202322C2780451532 @default.
- W2981202322 hasConceptScore W2981202322C37736160 @default.
- W2981202322 hasConceptScore W2981202322C41008148 @default.
- W2981202322 hasConceptScore W2981202322C55493867 @default.
- W2981202322 hasConceptScore W2981202322C63479239 @default.
- W2981202322 hasConceptScore W2981202322C98045186 @default.
- W2981202322 hasLocation W29812023221 @default.
- W2981202322 hasOpenAccess W2981202322 @default.
- W2981202322 hasPrimaryLocation W29812023221 @default.
- W2981202322 hasRelatedWork W2555661914 @default.
- W2981202322 hasRelatedWork W2754251838 @default.
- W2981202322 hasRelatedWork W2759816466 @default.
- W2981202322 hasRelatedWork W2790777763 @default.
- W2981202322 hasRelatedWork W2891902321 @default.
- W2981202322 hasRelatedWork W2904093354 @default.
- W2981202322 hasRelatedWork W2918566147 @default.
- W2981202322 hasRelatedWork W2944668088 @default.