{"id":"W3197873762","doi":"10.1038/s41598-022-07167-7","title":"A comparison of approaches to improve worst-case predictive model performance over patient subpopulations","year":2022,"lang":"en","type":"preprint","venue":"Scientific Reports","topic":"Machine Learning in Healthcare","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"National Heart, Lung, and Blood Institute; Stanford Research Computing Center, Stanford University; National Science Foundation","keywords":"Metric (unit); Computer science; Machine learning; Performance metric; Scale (ratio); Population; Predictive modelling; Predictive power; Quality (philosophy); Empirical research; Artificial intelligence; Medicine; Statistics; Mathematics; Operations management","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":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.002052779,0.0003782475,0.0006273469,0.0006646615,0.00081128,0.0004117443,0.001037285,0.000185238,0.00004793596],"category_scores_gemma":[0.0002758122,0.0003869009,0.0002205519,0.001050522,0.0001470717,0.0003108573,0.005073305,0.001206479,0.00000375999],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003983427,"about_ca_system_score_gemma":0.0009464316,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003867101,"about_ca_topic_score_gemma":0.00007750678,"domain_scores_codex":[0.9936911,0.000240051,0.001559073,0.002356678,0.001618158,0.0005348921],"domain_scores_gemma":[0.9935887,0.00007350813,0.001727093,0.003985748,0.0003153718,0.0003095779],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00001013157,0.0001981933,0.07721204,0.0003230011,0.00002307751,0.0001994286,0.01331544,0.8988606,0.00003347249,0.0004601438,0.001355567,0.008008897],"study_design_scores_gemma":[0.00004872985,0.0001613439,0.003795477,0.00009717089,0.00002081259,0.0001880343,0.0001822039,0.9905986,0.0003673197,0.003462076,0.0007294737,0.0003488113],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8787588,0.0001379536,0.1070613,0.0002567295,0.01078626,0.001829154,0.00004849521,0.0002256106,0.0008956918],"genre_scores_gemma":[0.9617952,6.382816e-7,0.03681633,0.00002465903,0.00004675573,0.0005329911,0.0001365678,0.00002993224,0.0006168993],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.09173794,"threshold_uncertainty_score":0.9998583,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1232113807884043,"score_gpt":0.3371047260080542,"score_spread":0.2138933452196499,"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."}}