{"id":"W3206966921","doi":"10.48550/arxiv.2110.05524","title":"Generalization Techniques Empirically Outperform Differential Privacy against Membership Inference","year":2021,"lang":"en","type":"preprint","venue":"arXiv (Cornell University)","topic":"Privacy-Preserving Technologies in Data","field":"Computer Science","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"","keywords":"Differential privacy; Generalization; Inference; Computer science; Differential (mechanical device); Privacy protection; Computer security; Data mining; Artificial intelligence; Internet privacy; Machine learning; Mathematics; Engineering","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":[0.01713733,0.001577451,0.001824307,0.001172708,0.001768481,0.003164876,0.003350525,0.003773706,0.002817616],"category_scores_gemma":[0.07526447,0.0006575286,0.001904692,0.001570292,0.004646878,0.008341796,0.006711389,0.006874741,0.0009486959],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003314772,"about_ca_system_score_gemma":0.002288198,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008739237,"about_ca_topic_score_gemma":0.0009483977,"domain_scores_codex":[0.9843563,0.00699129,0.0006713363,0.002631961,0.004123374,0.001225669],"domain_scores_gemma":[0.92755,0.04178172,0.003741173,0.02352424,0.002409064,0.0009938007],"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":[0.001609945,0.0006706293,0.01242153,0.0005000761,0.0004705986,0.0002860107,0.0006830567,0.476917,0.01336569,0.2731984,0.007359636,0.2125175],"study_design_scores_gemma":[0.0001040132,0.0003874656,0.001299344,0.00007673926,0.00007562622,0.0004144392,0.0001155323,0.7702376,0.02098096,0.2029417,0.003323832,0.00004279933],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1513676,0.001606531,0.8296668,0.004339697,0.0001370969,0.0001712394,0.0003463244,0.002082739,0.01028197],"genre_scores_gemma":[0.888711,0.0004073561,0.107245,0.0007321705,0.0000899739,0.0001072669,0.0002676719,0.0001969402,0.002242554],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01713733,"threshold_uncertainty_score":0.09063196,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1103616220539324,"score_gpt":0.2419881294932673,"score_spread":0.131626507439335,"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."}}