{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001353275,0.001648578,0.0006215646,0.0009567377,0.0004161506,0.0005415002,0.001622568,0.000904542,0.001509198],"category_scores_gemma":[0.004245228,0.0005074716,0.0006834724,0.0003230411,0.0008885933,0.001313669,0.001141113,0.001539121,0.0004473212],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001388564,"about_ca_system_score_gemma":0.001080606,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005636518,"about_ca_topic_score_gemma":0.005879547,"domain_scores_codex":[0.9994767,0.00007364654,0.0000344773,0.0001692783,0.0001712789,0.0000745608],"domain_scores_gemma":[0.9986402,0.0005035862,0.0002113728,0.0002295224,0.0003697811,0.00004549377],"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":[0.0001965077,0.00009052072,0.00237949,0.00009470807,0.00007516816,0.0002667471,0.0001425484,0.748163,0.02564556,0.003738632,0.00136727,0.2178399],"study_design_scores_gemma":[0.00000261638,0.00003297408,0.0003477883,0.000007517243,0.000008073981,0.00002746258,0.00000872413,0.9868024,0.01072814,0.001732154,0.0002964691,0.000005828575],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.06660454,0.0002599859,0.9285427,0.0001518763,0.00005835583,0.00008912724,0.00007449005,0.002699926,0.001518975],"genre_scores_gemma":[0.7867711,0.0001932821,0.2092988,0.0001975359,0.00002412646,0.0001138187,0.0002897043,0.0002907994,0.002820884],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.005636518,"threshold_uncertainty_score":0.0112074,"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."}}