{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0009482648,0.0001867059,0.0002177986,0.0003739025,0.0002676194,0.0004619776,0.001036334,0.0001023317,0.00004438575],"category_scores_gemma":[0.001910962,0.000181535,0.0001263482,0.0008795336,0.00004892598,0.0006801325,0.0005690144,0.0004543997,0.00002845024],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001779746,"about_ca_system_score_gemma":0.0001578391,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001402356,"about_ca_topic_score_gemma":0.0003239818,"domain_scores_codex":[0.9980923,0.00008947946,0.0003883305,0.0006755336,0.0002652628,0.0004891171],"domain_scores_gemma":[0.9974755,0.001714971,0.00005030673,0.000603439,0.00005887189,0.00009690781],"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":[0.00001889433,0.0000489561,0.00176934,0.0001138732,0.00005787745,0.0002242477,0.002849084,0.3082648,0.000882685,0.1843957,0.002141655,0.4992329],"study_design_scores_gemma":[0.0004171069,0.00002045302,0.0004824842,0.0001784505,0.00000777559,0.00003524757,0.0004065807,0.9833946,0.0001645026,0.01252946,0.002119201,0.0002441105],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.005168076,0.0004607386,0.9853896,0.001822466,0.001340994,0.0002674446,0.000001228122,0.0004893836,0.005060013],"genre_scores_gemma":[0.6046171,0.000006684662,0.3945795,0.0001735927,0.00009833591,0.00003265456,0.000001186332,0.00002328418,0.0004677106],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.6751298,"threshold_uncertainty_score":0.7402782,"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."}}