{"id":"W4389523936","doi":"10.18653/v1/2023.emnlp-main.265","title":"Preserving Privacy Through Dememorization: An Unlearning Technique For Mitigating Memorization Risks In Language Models","year":2023,"lang":"en","type":"article","venue":"","topic":"Privacy-Preserving Technologies in Data","field":"Computer Science","cited_by":10,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Windsor","funders":"Natural Sciences and Engineering Research Council of Canada; Alliance de recherche numérique du Canada","keywords":"Memorization; Computer science; Reinforcement learning; Replication (statistics); Artificial intelligence; Similarity (geometry); Quality (philosophy); Cognitive psychology; Psychology","routes":{"ca_aff":true,"ca_fund":true,"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":["metaresearch","open_science"],"consensus_categories":["open_science"],"category_scores_codex":[0.001145767,0.0002062328,0.0002184495,0.0003375041,0.0002388441,0.0002568592,0.01417469,0.0002064064,0.00001132083],"category_scores_gemma":[0.01312862,0.0002141921,0.00004424896,0.002309077,0.00004125931,0.003998247,0.02894852,0.0002901848,0.000009372559],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000962611,"about_ca_system_score_gemma":0.00007201242,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002704032,"about_ca_topic_score_gemma":0.00003690587,"domain_scores_codex":[0.9977726,0.0001309676,0.0004402688,0.0007800491,0.000332482,0.0005436382],"domain_scores_gemma":[0.9946271,0.0003373638,0.0001728094,0.004701912,0.0001103133,0.00005049155],"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":[0.00005197185,0.0004758188,0.02094426,0.001194437,0.0001264564,0.0001899508,0.02787896,0.2188595,0.1983282,0.346599,0.06048448,0.1248669],"study_design_scores_gemma":[0.000196427,0.00003680658,0.0002913691,0.00006446362,0.000002207149,0.000003146222,0.0003059015,0.6655971,0.03486186,0.2983671,0.00008729726,0.0001863237],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.02977052,0.00003678601,0.959857,0.004568807,0.0001647883,0.0009612325,0.000009453806,0.003504192,0.00112717],"genre_scores_gemma":[0.3839126,0.00002306399,0.6154494,0.00007903774,0.00004950111,0.0003038132,0.0001001138,0.00002701972,0.00005543248],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.4467376,"threshold_uncertainty_score":0.9951842,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09904560295695426,"score_gpt":0.3578418422760816,"score_spread":0.2587962393191273,"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."}}