{"id":"W6891809573","doi":"10.48448/fjt5-8369","title":"Rethinking Label Refurbishment: Model Robustness under Label Noise","year":2023,"lang":"en","type":"other","venue":"Underline Science Inc.","topic":"","field":"","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University","funders":"","keywords":"Robustness (evolution); Generalization; Artificial neural network; Regularization (linguistics); Deep neural networks; Noise (video); Pattern recognition (psychology)","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":[0.006992847,0.001851802,0.00128725,0.0008478412,0.001038605,0.002170325,0.003905879,0.002881213,0.00178905],"category_scores_gemma":[0.02611605,0.0007184877,0.001302006,0.0006761679,0.002785381,0.004143894,0.003898154,0.005232024,0.0008334792],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001540426,"about_ca_system_score_gemma":0.001605982,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007932258,"about_ca_topic_score_gemma":0.008322604,"domain_scores_codex":[0.9971935,0.001195337,0.0001062611,0.0008626537,0.0004407034,0.0002015869],"domain_scores_gemma":[0.9885939,0.005758239,0.001213566,0.003230445,0.0008655876,0.0003381846],"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":[0.0008936867,0.0002973152,0.01240702,0.0002554022,0.000277776,0.0003036769,0.0006979251,0.7184733,0.01173519,0.03648373,0.008088904,0.210086],"study_design_scores_gemma":[0.00002795165,0.00004039975,0.0004271986,0.00002218686,0.00001602409,0.00004405597,0.00002288128,0.9739912,0.002354864,0.02220095,0.0008325081,0.00001972285],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1034944,0.001339044,0.8870496,0.002265292,0.0001817126,0.0001043466,0.000313346,0.002787957,0.002464333],"genre_scores_gemma":[0.8085393,0.0005129628,0.1835995,0.00130828,0.0002141949,0.0001616892,0.0008293096,0.0007591068,0.00407561],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.007932258,"threshold_uncertainty_score":0.03698218,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08127156378226208,"score_gpt":0.3313697929390731,"score_spread":0.250098229156811,"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."}}