{"id":"W4404782635","doi":"10.18653/v1/2024.emnlp-main.537","title":"Attribute Diversity Determines the Systematicity Gap in VQA","year":2024,"lang":"en","type":"article","venue":"","topic":"Quality Function Deployment in Product Design","field":"Business, Management and Accounting","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Samsung; University of Toronto; Government of Canada; Canadian Institute for Advanced Research","keywords":"Diversity (politics); Computer science; Political science","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01381816,0.001129262,0.001101179,0.001055484,0.00101755,0.002809795,0.001793224,0.002075678,0.003985392],"category_scores_gemma":[0.06617629,0.0006626807,0.001185302,0.001062386,0.002325193,0.007210092,0.003689318,0.004102194,0.001243884],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001716381,"about_ca_system_score_gemma":0.00167536,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004300409,"about_ca_topic_score_gemma":0.005066691,"domain_scores_codex":[0.99103,0.004552468,0.0005705482,0.00214397,0.001315062,0.0003881197],"domain_scores_gemma":[0.9386467,0.0473582,0.001662269,0.009402996,0.002182739,0.0007470174],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00335033,0.0008522407,0.08630203,0.002153893,0.0007278702,0.0007775229,0.005572305,0.1520429,0.0351457,0.05353142,0.03460148,0.6249424],"study_design_scores_gemma":[0.0003948711,0.0008817738,0.02268203,0.0003554455,0.0002127157,0.001017926,0.001435387,0.6719217,0.0223946,0.2588133,0.01976633,0.0001239639],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6827803,0.006502217,0.2783347,0.003530161,0.0002607968,0.000324457,0.00393745,0.006745955,0.017584],"genre_scores_gemma":[0.943092,0.0004023595,0.05028701,0.0005600996,0.00006227259,0.0001027649,0.004039214,0.0004155469,0.001038819],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01381816,"threshold_uncertainty_score":0.07307827,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09213335946445604,"score_gpt":0.2596470452020815,"score_spread":0.1675136857376254,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}