{"id":"W7114916806","doi":"10.1162/tacl.a.62","title":"Towards More Realistic Extraction Attacks: An Adversarial Perspective","year":2025,"lang":"en","type":"article","venue":"Transactions of the Association for Computational Linguistics","topic":"Adversarial Robustness in Machine Learning","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Mila - Quebec Artificial Intelligence Institute","funders":"","keywords":"Adversarial system; Adversary; Perspective (graphical); Training set; Memorization; Attack model","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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0005215677,0.0001374707,0.0001945677,0.0001661746,0.0005325443,0.00008026137,0.000636186,0.0001252481,0.000005143613],"category_scores_gemma":[0.003766231,0.0001335548,0.0001960838,0.0005812046,0.00005234641,0.0001684474,0.00003149633,0.0002625004,0.000001560548],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007902506,"about_ca_system_score_gemma":0.0004632347,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002179825,"about_ca_topic_score_gemma":0.00003003273,"domain_scores_codex":[0.9985347,0.0001437909,0.0003839767,0.0002928427,0.0004652686,0.0001793743],"domain_scores_gemma":[0.9962007,0.0009275091,0.0004253004,0.000308413,0.002099644,0.00003840398],"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":[0.00002397498,0.00008025417,0.0001952496,0.00001384692,0.00008302106,1.238246e-7,0.000433967,0.6246652,0.000007860745,0.3736567,0.0001533554,0.0006863797],"study_design_scores_gemma":[0.0008216348,0.00005974794,0.00761136,0.00003115757,0.0001617074,9.735311e-7,0.0002733322,0.8221319,0.0001070688,0.165831,0.002819275,0.0001508175],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.0002345632,0.00001339677,0.9902978,0.001832291,0.003384445,0.0003283721,0.00008495159,0.0001195697,0.003704646],"genre_scores_gemma":[0.8620734,0.000001713251,0.1366552,0.0001266174,0.0002637356,0.00002111933,0.00002840419,0.00001090654,0.000818901],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.8618389,"threshold_uncertainty_score":0.5446206,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01430245061515127,"score_gpt":0.3394118539592187,"score_spread":0.3251094033440675,"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."}}