{"id":"W4404783705","doi":"10.18653/v1/2024.emnlp-main.211","title":"Decompose and Compare Consistency: Measuring VLMs’ Answer Reliability via Task-Decomposition Consistency Comparison","year":2024,"lang":"en","type":"article","venue":"","topic":"Topic Modeling","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Canadian Institute for Advanced Research","keywords":"Consistency (knowledge bases); Reliability (semiconductor); Computer science; Task (project management); Decomposition; Reliability engineering; Data mining; Artificial intelligence; Engineering","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":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0006985977,0.0002928464,0.0004417653,0.0001744041,0.0003048252,0.0005804913,0.0005199809,0.00009332669,0.00004845862],"category_scores_gemma":[0.00004181189,0.0002683703,0.000122974,0.0003342491,0.0001511347,0.0007667842,0.000380367,0.000331262,0.00008692477],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001313253,"about_ca_system_score_gemma":0.0001331621,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002217446,"about_ca_topic_score_gemma":0.0001213211,"domain_scores_codex":[0.9972729,0.000171441,0.0006972565,0.001019577,0.0004208153,0.0004179765],"domain_scores_gemma":[0.9983095,0.0003476698,0.00007812851,0.0008703111,0.0001687448,0.0002256778],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00009952653,0.001189948,0.1062972,0.002313688,0.0005209427,0.0004694917,0.0135472,0.007841759,0.03019081,0.475848,0.01030182,0.3513796],"study_design_scores_gemma":[0.0003308812,0.00006633986,0.006076104,0.0001903434,0.00003581982,0.0001581503,0.000114062,0.975932,0.001681644,0.01182775,0.003123677,0.000463217],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1830076,0.002509435,0.8022377,0.002146323,0.001179439,0.0003534767,0.00000472383,0.0009911238,0.00757013],"genre_scores_gemma":[0.90389,0.00001936701,0.09559771,0.0002772702,0.00007377454,0.00001690584,0.000006575143,0.00001614494,0.0001023129],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9680902,"threshold_uncertainty_score":0.9999769,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03196609638306724,"score_gpt":0.2864189849729321,"score_spread":0.2544528885898649,"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."}}