{"id":"W4391444574","doi":"10.1145/3641543","title":"Beyond Fidelity: Explaining Vulnerability Localization of Learning-Based Detectors","year":2024,"lang":"en","type":"article","venue":"ACM Transactions on Software Engineering and Methodology","topic":"Adversarial Robustness in Machine Learning","field":"Computer Science","cited_by":9,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Computer science; Fidelity; Vulnerability (computing); Detector; Computer security; Telecommunications","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004223689,0.001672167,0.0008915964,0.003258075,0.0004452583,0.00158543,0.001482202,0.001840652,0.001465975],"category_scores_gemma":[0.0321026,0.0004462941,0.001015641,0.001333087,0.001590216,0.00477336,0.00236011,0.002482026,0.0004319641],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001220195,"about_ca_system_score_gemma":0.001315052,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003842303,"about_ca_topic_score_gemma":0.003898358,"domain_scores_codex":[0.9976934,0.0007747994,0.000149778,0.0005247498,0.0006296025,0.0002277313],"domain_scores_gemma":[0.97497,0.01690649,0.002942454,0.003280211,0.001596535,0.0003041533],"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":[0.0003909981,0.0002019511,0.05609391,0.0004294323,0.0002831156,0.0005466797,0.0006159511,0.6625903,0.007769792,0.02146492,0.004670677,0.2449424],"study_design_scores_gemma":[0.00001381302,0.0000731121,0.002633271,0.00003930067,0.00004627704,0.0001752034,0.00007783039,0.9688626,0.005071422,0.02183164,0.001149256,0.00002633956],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1543807,0.0009353022,0.8363457,0.001175744,0.00007136853,0.0001253228,0.0005656753,0.00441859,0.001981409],"genre_scores_gemma":[0.8890606,0.0003041814,0.10843,0.0002604788,0.00003953441,0.00005189999,0.0008515617,0.0001994009,0.0008024033],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.004223689,"threshold_uncertainty_score":0.02233726,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04447907595058553,"score_gpt":0.3148810260823837,"score_spread":0.2704019501317981,"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."}}