{"id":"W3134750715","doi":"10.1158/1557-3265.adi21-ia-15","title":"Abstract IA-15: Platforms to improve reproducibility in artificial intelligence research","year":2021,"lang":"en","type":"article","venue":"Clinical Cancer Research","topic":"Radiomics and Machine Learning in Medical Imaging","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto; Princess Margaret Cancer Centre","funders":"","keywords":"Reproducibility; Computer science; Transparency (behavior); Artificial intelligence; Field (mathematics); Data science; Mathematics","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1151992,0.001720523,0.002115478,0.00615328,0.003288186,0.01827772,0.006616383,0.004439121,0.02291156],"category_scores_gemma":[0.1769492,0.00191004,0.003065344,0.003727997,0.005132579,0.0148192,0.03096453,0.005911086,0.01593283],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003830785,"about_ca_system_score_gemma":0.0118694,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001963785,"about_ca_topic_score_gemma":0.0009584374,"domain_scores_codex":[0.9416904,0.02973404,0.005915501,0.00605384,0.01365996,0.002946337],"domain_scores_gemma":[0.7318447,0.08370803,0.01802632,0.0915665,0.05590712,0.01894737],"domain_codex":null,"domain_gemma":"reproducibility","domain_candidate":"reproducibility","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.004184797,0.0009048032,0.01359332,0.001787113,0.0005918388,0.000987696,0.00548164,0.007480465,0.02293304,0.3337224,0.1815371,0.4267958],"study_design_scores_gemma":[0.001183165,0.001294576,0.01062188,0.0020872,0.000418226,0.0007332672,0.001634201,0.03184862,0.03702468,0.3342694,0.5782475,0.0006372027],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.03328427,0.004294594,0.7657272,0.04302714,0.00704567,0.004692604,0.007777543,0.04676026,0.08739073],"genre_scores_gemma":[0.2012293,0.001812514,0.7383425,0.005046995,0.002973107,0.007123481,0.01038032,0.007427156,0.02566456],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.8848008,"threshold_uncertainty_score":0.609239,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3980621339270541,"score_gpt":0.6218056127994603,"score_spread":0.2237434788724063,"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."}}