{"id":"W4378942317","doi":"10.48550/arxiv.2305.18393","title":"Training Private Models That Know What They Don't Know","year":2023,"lang":"en","type":"preprint","venue":"arXiv (Cornell University)","topic":"Privacy-Preserving Technologies in Data","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Natural Sciences and Engineering Research Council of Canada; Defense Advanced Research Projects Agency; Government of Canada; Canadian Institute for Advanced Research; Alfred P. Sloan Foundation","keywords":"Differential privacy; Harm; Computer science; Private information retrieval; Machine learning; Randomness; Empirical research; Artificial intelligence; Process (computing); Computer security; Data mining","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01568515,0.00135712,0.001968335,0.0006687903,0.00129453,0.004385169,0.003338922,0.003572182,0.002163977],"category_scores_gemma":[0.0656095,0.001127346,0.001504793,0.001170562,0.004183593,0.01500433,0.00515354,0.007813686,0.001603492],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002105847,"about_ca_system_score_gemma":0.002622677,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001063638,"about_ca_topic_score_gemma":0.001316797,"domain_scores_codex":[0.988515,0.004599112,0.0004985019,0.002284783,0.003083428,0.001019331],"domain_scores_gemma":[0.9286826,0.02740678,0.0041827,0.03664851,0.002105114,0.000974296],"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":[0.002151275,0.0006811747,0.03147896,0.0005343477,0.0006248147,0.0006391726,0.001353798,0.479443,0.01774637,0.2124614,0.01001064,0.2428749],"study_design_scores_gemma":[0.00005898562,0.000239057,0.00176224,0.00007713334,0.00008480114,0.000374118,0.0001983191,0.730872,0.01918819,0.2434656,0.003632101,0.00004741581],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.1406295,0.0005678168,0.846437,0.005295803,0.0001049741,0.0001235664,0.0006278797,0.001250673,0.00496261],"genre_scores_gemma":[0.9220817,0.0003785563,0.0738987,0.0007989031,0.00009148107,0.000131159,0.0005013882,0.0002023503,0.001915832],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.01568515,"threshold_uncertainty_score":0.08295202,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2268399420354769,"score_gpt":0.2333432327983583,"score_spread":0.006503290762881336,"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."}}