{"id":"W4400104927","doi":"10.52202/079017-0733","title":"Detecting Brittle Decisions for Free: Leveraging Margin Consistency in Deep Robust Classifiers","year":2024,"lang":"en","type":"article","venue":"","topic":"Adversarial Robustness in Machine Learning","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"Canadian Institute for Advanced Research; Université Laval; Mila - Quebec Artificial Intelligence Institute","funders":"Natural Sciences and Engineering Research Council of Canada; Consortium de Recherche et d’innovation en Aérospatiale au Québec","keywords":"Margin (machine learning); Computer science; Robustness (evolution); Consistency (knowledge bases); Logit; Artificial intelligence; Machine learning; Adversarial system","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":[0.005269276,0.001661122,0.001420623,0.001153523,0.000591956,0.001954745,0.002159134,0.00177527,0.001395864],"category_scores_gemma":[0.03022158,0.000789155,0.0008360147,0.0006052773,0.002668295,0.00494588,0.004304991,0.004415447,0.0005364494],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001171271,"about_ca_system_score_gemma":0.001074357,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001372722,"about_ca_topic_score_gemma":0.001426485,"domain_scores_codex":[0.9974365,0.0008457397,0.0001552282,0.0005741013,0.0007213659,0.0002672013],"domain_scores_gemma":[0.9857949,0.007450905,0.002497476,0.002841352,0.0009981587,0.0004171561],"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":[0.0003620613,0.0001211846,0.007396553,0.00009333631,0.0001327412,0.0002388348,0.0001872158,0.8792823,0.00672511,0.01468176,0.002090279,0.08868872],"study_design_scores_gemma":[0.000005949975,0.00005819701,0.0004352057,0.00001686652,0.00001221573,0.00005048848,0.00001766607,0.9784518,0.002899919,0.01778466,0.0002526305,0.00001442044],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1169705,0.0004612711,0.8782433,0.0008421571,0.00005248587,0.00007145706,0.0001315504,0.001562698,0.001664369],"genre_scores_gemma":[0.9505824,0.0001438555,0.0475622,0.0003406725,0.00004977055,0.00005593468,0.0001952847,0.0001651856,0.000904728],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.005269276,"threshold_uncertainty_score":0.0278669,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0437626189196848,"score_gpt":0.2856479761025212,"score_spread":0.2418853571828364,"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."}}