{"id":"W3173445857","doi":"10.1101/2021.06.17.21259092","title":"Evaluation of Domain Generalization and Adaptation on Improving Model Robustness to Temporal Dataset Shift in Clinical Medicine","year":2021,"lang":"en","type":"preprint","venue":"medRxiv","topic":"Machine Learning in Healthcare","field":"Computer Science","cited_by":10,"is_retracted":false,"has_abstract":true,"ca_institutions":"Institute for Clinical Evaluative Sciences; SickKids Foundation; Hospital for Sick Children","funders":"","keywords":"Computer science; Artificial intelligence; Robustness (evolution); Receiver operating characteristic; False positive paradox; Machine learning; Predictive modelling; Data mining","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":[],"consensus_categories":[],"category_scores_codex":[0.01233044,0.0002052472,0.0004428845,0.0003318625,0.00004675302,0.00005592127,0.0005751975,0.0002352025,0.000007746768],"category_scores_gemma":[0.001830152,0.0001963902,0.0000339222,0.0003492764,0.00004362188,0.0001583821,0.0007847673,0.0005829782,6.812807e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001342237,"about_ca_system_score_gemma":0.000636475,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001192488,"about_ca_topic_score_gemma":0.001810907,"domain_scores_codex":[0.9946445,0.002005363,0.001076048,0.0009118866,0.001176805,0.0001853621],"domain_scores_gemma":[0.9977527,0.0002076661,0.000532206,0.0009921531,0.0003920886,0.000123182],"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":[0.00001212634,0.00004060218,0.02925409,0.0002037632,0.00000751379,0.000004801093,0.001904674,0.9186525,0.00002467746,0.0009127767,0.0000398403,0.04894258],"study_design_scores_gemma":[0.0004498021,0.00009756636,0.06710182,0.0003791977,0.00002210549,0.000001439902,0.00004744146,0.9297963,0.000008811274,0.001939713,0.00000775513,0.0001480942],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5121982,0.0001000185,0.4851882,0.00170967,0.0003386908,0.0004080107,0.00002634701,0.00002020904,0.00001070243],"genre_scores_gemma":[0.9028989,0.00002034237,0.09559959,0.0002337386,0.0001181551,0.00006427442,0.00104807,0.00001500955,0.000001924581],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3907008,"threshold_uncertainty_score":0.8008561,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1508825085652468,"score_gpt":0.4134894985176525,"score_spread":0.2626069899524057,"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."}}