{"id":"W4403795826","doi":"10.48550/arxiv.2409.17538","title":"LoRA Provides Differential Privacy by Design via Random Sketching","year":2024,"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":"University of Waterloo; DeepMind; Canadian Institute for Advanced Research","keywords":"Adaptation (eye); Relation (database); Rank (graph theory); Differential privacy; Differential (mechanical device); Computer science; Internet privacy; Computer security; Psychology; Mathematics; Data mining; Physics; Thermodynamics; Combinatorics","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":[0.007371695,0.00166444,0.001999111,0.0007520785,0.0009786861,0.003026542,0.003101165,0.002365369,0.003899728],"category_scores_gemma":[0.03572248,0.001339586,0.001567339,0.0008923149,0.003482717,0.00602554,0.00759255,0.004647805,0.002052422],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001321905,"about_ca_system_score_gemma":0.001974973,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0006142269,"about_ca_topic_score_gemma":0.0007146081,"domain_scores_codex":[0.991463,0.004271818,0.0003961494,0.001571396,0.001644298,0.0006533117],"domain_scores_gemma":[0.9739277,0.01250019,0.001408315,0.01042267,0.001100661,0.0006404233],"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":[0.0007001408,0.000183134,0.002128617,0.0003714399,0.0002217813,0.0003761943,0.0005314061,0.4958878,0.0137974,0.3529046,0.005971969,0.1269255],"study_design_scores_gemma":[0.00005231116,0.0001180364,0.0001238319,0.00002559667,0.00002412776,0.0001527862,0.00002519161,0.8554068,0.003879395,0.1379683,0.002190005,0.00003357189],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.007892779,0.0001769856,0.9889319,0.0003401772,0.00002470771,0.00006490027,0.0001087819,0.001074529,0.001385188],"genre_scores_gemma":[0.7624162,0.0004500113,0.2304822,0.0006348507,0.0001274924,0.0004863502,0.0003089915,0.0004042673,0.004689547],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.007371695,"threshold_uncertainty_score":0.03898573,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06803968863021041,"score_gpt":0.2083793399622571,"score_spread":0.1403396513320467,"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."}}