{"id":"W4387389875","doi":"10.48550/arxiv.2310.02567","title":"Improving Automatic VQA Evaluation Using Large Language Models","year":2023,"lang":"en","type":"preprint","venue":"arXiv (Cornell University)","topic":"Multimodal Machine Learning Applications","field":"Computer Science","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Canadian Institute for Advanced Research; Samsung; Nvidia","keywords":"Leverage (statistics); Computer science; Metric (unit); Machine learning; Proxy (statistics); Question answering; Task (project management); Artificial intelligence; Set (abstract data type); Context (archaeology); Data mining; Information retrieval","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":[0.01851711,0.002051185,0.001501094,0.002923726,0.001116268,0.004121637,0.002757913,0.002330797,0.006831951],"category_scores_gemma":[0.08568747,0.0006313966,0.00150768,0.001271084,0.0009446856,0.005575601,0.004159186,0.002845657,0.003744698],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002185546,"about_ca_system_score_gemma":0.002317848,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.009747172,"about_ca_topic_score_gemma":0.01153384,"domain_scores_codex":[0.9670649,0.02175007,0.001884125,0.003214166,0.005335288,0.0007515167],"domain_scores_gemma":[0.934461,0.04151241,0.001928996,0.006909536,0.01403069,0.001157352],"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":[0.00166056,0.001062452,0.01849748,0.001736599,0.0005648854,0.0003949812,0.002400031,0.09604312,0.04148013,0.01182736,0.09356888,0.7307635],"study_design_scores_gemma":[0.0002566919,0.0005429888,0.005034442,0.0001641884,0.0001015312,0.0002276872,0.0005850379,0.9382066,0.02374361,0.01293071,0.01805972,0.0001468821],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2393579,0.006482032,0.6380796,0.002641955,0.001298874,0.001178362,0.005204452,0.08938675,0.01637002],"genre_scores_gemma":[0.7915302,0.0003904658,0.1884934,0.001056193,0.0002356896,0.0005712751,0.01049135,0.002627028,0.004604389],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01851711,"threshold_uncertainty_score":0.097929,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1362128373710949,"score_gpt":0.2611552107093333,"score_spread":0.1249423733382385,"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."}}