{"id":"W3135255132","doi":"","title":"Quantifying identifiability to choose and audit $\\epsilon$ in differentially private deep learning","year":2021,"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":"University of Waterloo","funders":"","keywords":"Differential privacy; Identifiability; Computer science; Adversary; Upper and lower bounds; Bounding overwatch; Inference; Bayesian inference; Audit; Data mining; Machine learning; Artificial intelligence; Computer security; Bayesian probability; 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":[],"consensus_categories":[],"category_scores_codex":[0.02836982,0.001652993,0.002041257,0.001401248,0.002480749,0.007345419,0.005045275,0.004626906,0.003385099],"category_scores_gemma":[0.1316643,0.001528152,0.001987346,0.00229289,0.01164624,0.01780647,0.01409651,0.01096889,0.0008105978],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.007426773,"about_ca_system_score_gemma":0.005271588,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001827985,"about_ca_topic_score_gemma":0.001828674,"domain_scores_codex":[0.9678661,0.01558872,0.001694439,0.005125035,0.007301636,0.002424003],"domain_scores_gemma":[0.881121,0.07998565,0.005835467,0.02770105,0.003969149,0.001387619],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0005685161,0.0001916509,0.004607428,0.0002361412,0.000121469,0.0003480286,0.001133753,0.1738968,0.003112483,0.7652076,0.002516967,0.04805918],"study_design_scores_gemma":[0.00003960507,0.00006515987,0.0003610715,0.00006867148,0.0000337424,0.0001460515,0.00009236266,0.3007202,0.00398502,0.6931024,0.001346585,0.00003908636],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.03836512,0.0003250818,0.9515037,0.004010595,0.00005881558,0.0001571214,0.0003534452,0.00044583,0.004780271],"genre_scores_gemma":[0.8530841,0.0005623469,0.1399791,0.001411396,0.0001295697,0.0005362494,0.0004685016,0.000228487,0.003600359],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.02836982,"threshold_uncertainty_score":0.1500358,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09038582693797159,"score_gpt":0.2308011545261982,"score_spread":0.1404153275882266,"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."}}