{"id":"W4367319323","doi":"10.3390/tomography9030074","title":"Can Machine Learning Be Better than Biased Readers?","year":2023,"lang":"en","type":"article","venue":"Tomography","topic":"COVID-19 diagnosis using AI","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"Nippon Steel Arts Foundation","keywords":"Computer science; Artificial intelligence; Machine learning; Binary classification; Regularization (linguistics); Classifier (UML); Convolutional neural network; Annotation; Baseline (sea); Natural language processing; Support vector machine","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04474036,0.001842705,0.002072052,0.001969771,0.001501338,0.006296785,0.002357735,0.005040887,0.005164095],"category_scores_gemma":[0.242073,0.0008340163,0.001236221,0.001786718,0.0051481,0.01027048,0.002873267,0.004025735,0.002951576],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002913037,"about_ca_system_score_gemma":0.002204783,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00318777,"about_ca_topic_score_gemma":0.002987376,"domain_scores_codex":[0.9712561,0.01572458,0.001497188,0.006422525,0.004216146,0.0008835129],"domain_scores_gemma":[0.8128223,0.1355861,0.01659711,0.01841606,0.01493018,0.001648158],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.001989191,0.0002978783,0.1923902,0.001693917,0.001482527,0.0008965925,0.002884882,0.04611791,0.00462528,0.06851563,0.03582591,0.6432802],"study_design_scores_gemma":[0.0003681383,0.001411224,0.07578344,0.002108985,0.000895519,0.002983333,0.001680271,0.292787,0.03089897,0.5149992,0.075488,0.0005960711],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2597628,0.02777851,0.5601141,0.1008269,0.003710853,0.0004685242,0.001660817,0.004546917,0.04113056],"genre_scores_gemma":[0.8802638,0.002552153,0.09136558,0.01597893,0.002106701,0.0002814298,0.0006650363,0.0008550229,0.005931233],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.04474036,"threshold_uncertainty_score":0.2366125,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04237204544308897,"score_gpt":0.304584667602852,"score_spread":0.262212622159763,"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."}}