{"id":"W4286485305","doi":"10.1145/3550271","title":"Black-box Safety Analysis and Retraining of DNNs based on Feature Extraction and Clustering","year":2022,"lang":"en","type":"article","venue":"ACM Transactions on Software Engineering and Methodology","topic":"Adversarial Robustness in Machine Learning","field":"Computer Science","cited_by":24,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Ottawa","funders":"Natural Sciences and Engineering Research Council of Canada; Fonds National de la Recherche Luxembourg; Université du Luxembourg","keywords":"Computer science; Cluster analysis; Retraining; Black box; Deep neural networks; Artificial intelligence; Machine learning; Artificial neural network; Feature (linguistics); Root (linguistics); Root cause; Pattern recognition (psychology); Reliability engineering","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":[0.001109311,0.0001609651,0.0003258088,0.0005673286,0.0002842367,0.00002263105,0.0002038658,0.00009556683,0.00002039156],"category_scores_gemma":[0.0005997974,0.0001759105,0.00006912712,0.0006579805,0.00005062703,0.0001186547,0.00004996445,0.0006688587,9.995073e-8],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004300999,"about_ca_system_score_gemma":0.00001911815,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002975433,"about_ca_topic_score_gemma":0.000006607719,"domain_scores_codex":[0.998678,0.0003770113,0.0001854216,0.0004111189,0.0001606081,0.0001878693],"domain_scores_gemma":[0.996607,0.002850167,0.00008395717,0.0003641183,0.00002352159,0.00007125441],"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":[0.00006143429,0.00001588953,0.0004244057,0.00003702837,0.0001080786,0.000006406145,0.001059947,0.9259316,0.0005137939,0.0001823322,0.000001410536,0.07165764],"study_design_scores_gemma":[0.0005023584,0.0002748998,0.0104453,0.00002162647,0.0001790339,0.00005080654,0.0002258781,0.9870867,0.0004639622,0.00007892297,0.00046033,0.0002101422],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.01357154,0.00006847094,0.9854473,0.000376655,0.0002787807,0.00007699127,0.00001214705,0.0001599577,0.000008102322],"genre_scores_gemma":[0.4433632,0.00002211445,0.5565005,0.00005689951,0.00001182381,0.000008739206,0.000002801335,0.00001006408,0.00002385781],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.4297916,"threshold_uncertainty_score":0.717342,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03701091279673228,"score_gpt":0.3033343891384405,"score_spread":0.2663234763417082,"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."}}