{"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":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001437003,0.0001233686,0.0001471944,0.0001793712,0.0002173958,0.0004763946,0.0002867079,0.00003701199,0.0005599482],"category_scores_gemma":[0.0001873941,0.00007739283,0.00006830829,0.000666903,0.00003834572,0.0009650015,0.000455681,0.0001169157,0.001616436],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004476035,"about_ca_system_score_gemma":0.000009060173,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007189969,"about_ca_topic_score_gemma":0.00019435,"domain_scores_codex":[0.9989604,0.00003400319,0.0002628898,0.0002600167,0.0002899019,0.0001928224],"domain_scores_gemma":[0.9994372,0.0001599083,0.00005554194,0.0002806699,0.00006130021,0.000005440006],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00005394457,0.0002142058,0.308916,0.00779476,0.0001884697,0.0001792939,0.0006334129,0.0003111669,0.0003829056,0.2921434,0.3854059,0.003776484],"study_design_scores_gemma":[0.002262043,0.00006119464,0.1863495,0.002847629,0.001026525,0.00006669428,0.006378876,0.157536,0.001200989,0.119746,0.519613,0.002911628],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9223321,0.0005307115,0.007802218,0.02257035,0.004646326,0.00139638,0.000004112331,0.001401497,0.03931631],"genre_scores_gemma":[0.9930366,0.000003628893,0.0000226766,0.00201656,0.0006964368,0.00001886441,0.000004422996,0.00001228421,0.004188573],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1723975,"threshold_uncertainty_score":0.9991609,"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."}}