{"id":"W3044427914","doi":"","title":"{UBCIS}: Ultimate Benchmark for Container Image Scanning","year":2020,"lang":"en","type":"article","venue":"USENIX Security Symposium","topic":"Medical Imaging Techniques and Applications","field":"Medicine","cited_by":11,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Waterloo","funders":"","keywords":"Container (type theory); Benchmark (surveying); Computer science; Image (mathematics); Computer vision; Artificial intelligence; Materials science; Geology; Composite material","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":[0.003125174,0.004478917,0.001691273,0.003532311,0.001443445,0.003643002,0.008601028,0.003138211,0.03141647],"category_scores_gemma":[0.0128023,0.001408479,0.001702292,0.004554241,0.0009872519,0.003487877,0.004771446,0.002265482,0.02314893],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001941881,"about_ca_system_score_gemma":0.003230397,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01895601,"about_ca_topic_score_gemma":0.02031114,"domain_scores_codex":[0.9957588,0.0006241275,0.0004667256,0.0007922102,0.001762342,0.0005957028],"domain_scores_gemma":[0.9939621,0.001318044,0.000210392,0.001740405,0.00221655,0.0005524505],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.002972491,0.0005739482,0.003483973,0.001375179,0.0003439992,0.0004063854,0.0002012647,0.012965,0.01194626,0.007065715,0.7540556,0.2046101],"study_design_scores_gemma":[0.001528825,0.00108243,0.00824558,0.0006613688,0.000296295,0.001127091,0.000598699,0.5028103,0.1361524,0.02471004,0.3224603,0.0003267357],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"software","genre_gemma":"empirical","genre_scores_codex":[0.07955965,0.004767654,0.1323594,0.001860016,0.001943818,0.001091271,0.07433756,0.64024,0.06384083],"genre_scores_gemma":[0.2491924,0.002078385,0.3283956,0.001966175,0.0002761466,0.001049551,0.3283637,0.05980399,0.02887403],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.03141647,"threshold_uncertainty_score":0.1050986,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02076700369625764,"score_gpt":0.3083486786194348,"score_spread":0.2875816749231772,"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."}}