{"id":"W4404986388","doi":"10.48550/arxiv.2411.15391","title":"Disk and Partial Disk Inspection: Worst- to Average-Case and Pareto Upper Bounds","year":2024,"lang":"en","type":"preprint","venue":"arXiv (Cornell University)","topic":"Industrial Vision Systems and Defect Detection","field":"Engineering","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Computer science; Business","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0001644394,0.0003098406,0.0003173682,0.0002994741,0.0001972396,0.0002429333,0.00008883156,0.0003872265,0.00003857525],"category_scores_gemma":[0.00002067795,0.0003355303,0.0001082912,0.0003673855,0.00005212409,0.0001115399,0.0004695635,0.0006677258,0.00008471831],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000195872,"about_ca_system_score_gemma":0.00002755857,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003584285,"about_ca_topic_score_gemma":0.0001775444,"domain_scores_codex":[0.9987655,0.00005439628,0.0002063019,0.0006579951,0.00006507854,0.0002507462],"domain_scores_gemma":[0.9992722,0.00004528432,0.00004053065,0.0003513317,0.0000413184,0.0002493708],"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":[0.0007654406,0.000135388,0.01576088,0.002487597,0.001626583,0.01736707,0.005792525,0.893346,0.0005677036,0.03024301,0.01300865,0.0188992],"study_design_scores_gemma":[0.002552162,0.0005058942,0.00328065,0.001382917,0.0009479676,0.001504252,0.002309049,0.9074895,0.001002164,0.01058463,0.0652933,0.003147546],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9842153,0.000277014,0.01013522,0.00002852918,0.002474717,0.0003858215,0.00006296211,0.0005780663,0.001842427],"genre_scores_gemma":[0.9983051,0.0001580856,0.00001636971,0.0000138357,0.0005370029,0.000004037429,0.000006260678,0.00004523597,0.0009140739],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.05228466,"threshold_uncertainty_score":0.9999097,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04558962420349932,"score_gpt":0.1807616159571679,"score_spread":0.1351719917536685,"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."}}