{"id":"W7125811146","doi":"10.21428/594757db.f2451ce9","title":"Faithful Perturbations and Evaluations for Post-Hoc Local Explanation Methods","year":2025,"lang":"en","type":"article","venue":"","topic":"Explainable Artificial Intelligence (XAI)","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"University of Alberta; Natural Sciences and Engineering Research Council of Canada; Canadian Institute for Advanced Research","keywords":"Fidelity; Noise (video); Perturbation (astronomy); Quality (philosophy); Distribution (mathematics); Worry","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01412882,0.001304282,0.0008823675,0.001057651,0.00079274,0.001594445,0.002773989,0.002421913,0.006039997],"category_scores_gemma":[0.06333753,0.0005171741,0.0008472129,0.0006101031,0.002336659,0.003787976,0.003379748,0.003703158,0.001034006],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002482522,"about_ca_system_score_gemma":0.001417481,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002714714,"about_ca_topic_score_gemma":0.003589444,"domain_scores_codex":[0.9909911,0.005265167,0.0003395062,0.001077887,0.00196163,0.0003646544],"domain_scores_gemma":[0.9620155,0.02760687,0.001171487,0.005786282,0.002725053,0.0006947903],"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":[0.001709399,0.0006164103,0.009844935,0.000475261,0.0002252279,0.0002423328,0.0006241492,0.6044546,0.008562607,0.0320113,0.008005392,0.3332284],"study_design_scores_gemma":[0.00008009739,0.0003614301,0.0008398014,0.00005160785,0.000021304,0.00005170692,0.0001122882,0.9765752,0.005522587,0.01475587,0.001608414,0.00001979842],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.1585578,0.001412619,0.8202319,0.001543375,0.0002855729,0.0008003559,0.0004234032,0.008224333,0.008520658],"genre_scores_gemma":[0.716735,0.0001736787,0.2785601,0.0003860921,0.00008918616,0.0003825056,0.0008870576,0.0006668268,0.002119585],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.01412882,"threshold_uncertainty_score":0.07472122,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03886993933859015,"score_gpt":0.4173731409240598,"score_spread":0.3785032015854697,"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."}}