{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02159141,0.001542912,0.001198253,0.001252022,0.0005108701,0.001491218,0.001613042,0.001751417,0.001058212],"category_scores_gemma":[0.04350753,0.0004250367,0.001615139,0.0007235535,0.0009234876,0.002040955,0.002043384,0.002449973,0.0004536462],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001185771,"about_ca_system_score_gemma":0.00158416,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006794909,"about_ca_topic_score_gemma":0.003685371,"domain_scores_codex":[0.9944451,0.003287297,0.0004261605,0.001232617,0.0003850289,0.0002237949],"domain_scores_gemma":[0.9733946,0.018073,0.001781654,0.00351998,0.002586493,0.000644211],"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":[0.001816286,0.0008915347,0.05286748,0.0002363377,0.0009634531,0.0001239567,0.0001916568,0.7953805,0.003480001,0.0007643302,0.003028315,0.1402561],"study_design_scores_gemma":[0.00007450329,0.000704185,0.008197919,0.00005229997,0.0001034242,0.00005798212,0.00007489968,0.9865737,0.002411961,0.001202137,0.0005165885,0.00003036414],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8613467,0.003566366,0.1255307,0.002341013,0.0004087728,0.0004035471,0.0008926872,0.002631439,0.002878751],"genre_scores_gemma":[0.9697063,0.0002274507,0.02789506,0.00039089,0.0000852189,0.00009091736,0.001077719,0.00006909719,0.0004574098],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02159141,"threshold_uncertainty_score":0.1141877,"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."}}