{"id":"W6969313300","doi":"10.5281/zenodo.8256421","title":"A Probabilistic Framework for Mutation Testing in Deep Neural Networks - Models archive Part 4","year":2022,"lang":"en","type":"article","venue":"PolyPublie (École Polytechnique de Montréal)","topic":"Generative Adversarial Networks and Image Synthesis","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Polytechnique Montréal","funders":"","keywords":"Probabilistic logic; Replication (statistics); Artificial neural network; Mutation; Software; Statistical model","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":[0.008584358,0.001456363,0.001515296,0.001934749,0.0007260598,0.002117382,0.003979133,0.002317356,0.009247274],"category_scores_gemma":[0.02596431,0.00136931,0.002384155,0.001401642,0.002484019,0.002715848,0.003840312,0.004783387,0.001816749],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002246394,"about_ca_system_score_gemma":0.003112945,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007467313,"about_ca_topic_score_gemma":0.01080932,"domain_scores_codex":[0.9950265,0.002318365,0.0002560944,0.0006642875,0.001380573,0.0003541693],"domain_scores_gemma":[0.9873563,0.009268898,0.0005726809,0.001206743,0.00126864,0.0003266955],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00009363041,0.00007435651,0.0007872277,0.0001046924,0.00009110785,0.0001208909,0.00007513842,0.7751819,0.001547973,0.1674993,0.005966768,0.048457],"study_design_scores_gemma":[0.000008248494,0.00001423991,0.00007651201,0.0000185346,0.000007990408,0.00002563225,0.000003461429,0.943958,0.0004229273,0.05439295,0.001061774,0.000009817662],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"dataset","genre_scores_codex":[0.001598946,0.000222176,0.9958163,0.0002632282,0.00003453264,0.00004843288,0.0001982826,0.0007024623,0.001115798],"genre_scores_gemma":[0.3475326,0.0008895983,0.6317973,0.0007987361,0.000431023,0.0009471704,0.002327932,0.002086679,0.01318895],"genre_candidate":"dataset","genre_consensus":null,"teacher_disagreement_score":0.009247274,"threshold_uncertainty_score":0.04539901,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02181960590637508,"score_gpt":0.2323912388658238,"score_spread":0.2105716329594487,"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."}}