{"id":"W3217026627","doi":"10.48550/arxiv.2111.15430","title":"The Devil is in the Margin: Margin-based Label Smoothing for Network Calibration","year":2021,"lang":"en","type":"preprint","venue":"arXiv (Cornell University)","topic":"Domain Adaptation and Few-Shot Learning","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Compute Canada","keywords":"Softmax function; Discriminative model; Margin (machine learning); Computer science; Artificial intelligence; Overfitting; Calibration; Machine learning; Artificial neural network; Mathematical optimization; Pattern recognition (psychology); Mathematics; Statistics","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003459951,0.001529474,0.001305158,0.00111672,0.001073732,0.001989685,0.00378192,0.0026676,0.005838247],"category_scores_gemma":[0.01557665,0.000721799,0.001174599,0.001231556,0.00227771,0.004771186,0.005594905,0.005190318,0.002577976],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001467937,"about_ca_system_score_gemma":0.001598414,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002351742,"about_ca_topic_score_gemma":0.003485302,"domain_scores_codex":[0.9981487,0.0005516062,0.00007960309,0.0006485721,0.0004226215,0.0001488992],"domain_scores_gemma":[0.9960052,0.001428511,0.0004400005,0.001450284,0.0004803964,0.0001955275],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0006323796,0.000227024,0.00298771,0.0002491847,0.0001494262,0.0002714717,0.0007407057,0.3803965,0.01969693,0.07408819,0.01787872,0.5026818],"study_design_scores_gemma":[0.00003375817,0.00005491567,0.0003545041,0.00005385052,0.00002473593,0.00008134793,0.00004645119,0.9298825,0.006880871,0.05753383,0.00501928,0.00003389444],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.01486725,0.0004469465,0.9780176,0.0007075771,0.0001195489,0.0000684377,0.0001392818,0.002979259,0.002654019],"genre_scores_gemma":[0.4869848,0.0005744904,0.4944046,0.001699266,0.0003366616,0.0004390547,0.0009840026,0.002486304,0.01209085],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.005838247,"threshold_uncertainty_score":0.01953089,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09479604473760554,"score_gpt":0.2060714126245323,"score_spread":0.1112753678869268,"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."}}