{"id":"W4416035385","doi":"10.18653/v1/2025.emnlp-main.1641","title":"REVIVING YOUR MNEME: Predicting The Side Effects of LLM Unlearning and Fine-Tuning via Sparse Model Diffing","year":2025,"lang":"","type":"article","venue":"","topic":"Explainable Artificial Intelligence (XAI)","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Alliance de recherche numérique du Canada; Canadian Institute for Advanced Research","keywords":"Action (physics); Feature (linguistics); Identification (biology); Perspective (graphical); Key (lock)","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":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","sts"],"consensus_categories":[],"category_scores_codex":[0.002567117,0.0005650913,0.0007364278,0.0004086584,0.001583463,0.000710876,0.001670241,0.0002191509,0.00001525906],"category_scores_gemma":[0.003601077,0.0004737563,0.0002073067,0.001836949,0.0003877631,0.001259528,0.002613356,0.001109487,0.00001409002],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000130824,"about_ca_system_score_gemma":0.0003747215,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001035331,"about_ca_topic_score_gemma":0.0002932715,"domain_scores_codex":[0.9950973,0.0005640296,0.001363669,0.001191458,0.0006421033,0.001141374],"domain_scores_gemma":[0.9941576,0.003303395,0.0006609427,0.001303499,0.0003809947,0.0001935976],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00004286992,0.0002372342,0.01108022,0.002213849,0.0002292182,0.0000659706,0.02243683,0.4056277,0.07329667,0.0671848,0.0001488239,0.4174359],"study_design_scores_gemma":[0.0001946166,0.0001478199,0.0009111031,0.002963158,0.0001259805,0.000011875,0.001433329,0.9304517,0.05790758,0.005414994,0.00005964441,0.0003782093],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2088285,0.003211425,0.7797523,0.001398753,0.0007318567,0.0007210969,6.593414e-7,0.0001551909,0.005200177],"genre_scores_gemma":[0.9782125,0.0003568098,0.01830823,0.0004666115,0.0001267102,0.00003200226,5.649013e-7,0.00003660009,0.002460015],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7693839,"threshold_uncertainty_score":0.9997714,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02425595569452285,"score_gpt":0.2749689856130517,"score_spread":0.2507130299185288,"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."}}