{"id":"W3173576323","doi":"10.48550/arxiv.2106.10865","title":"Benign Overfitting in Multiclass Classification: All Roads Lead to Interpolation","year":2021,"lang":"en","type":"preprint","venue":"arXiv (Cornell University)","topic":"Machine Learning and Data Classification","field":"Computer Science","cited_by":13,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Natural Sciences and Engineering Research Council of Canada; Simons Institute for the Theory of Computing, University of California Berkeley; King Abdullah University of Science and Technology; National Science Foundation","keywords":"Overfitting; Support vector machine; Artificial intelligence; Computer science; Interpretability; Machine learning; Pattern recognition (psychology); Multiclass classification; Structural risk minimization; Binary classification; Multinomial logistic regression; Artificial neural network; Mathematics; Algorithm","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":[0.01370222,0.001400664,0.00181332,0.001642436,0.001308104,0.002444308,0.002079117,0.002087237,0.001937052],"category_scores_gemma":[0.07540727,0.0008405889,0.001913877,0.001566567,0.005911302,0.005712049,0.005878891,0.006826777,0.0006053149],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001738719,"about_ca_system_score_gemma":0.001046801,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001289402,"about_ca_topic_score_gemma":0.0009145801,"domain_scores_codex":[0.9906228,0.004280353,0.0005291527,0.001680219,0.002375424,0.0005119877],"domain_scores_gemma":[0.96414,0.02001979,0.002504917,0.0103276,0.00235237,0.0006553821],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0003513485,0.0001059289,0.0062866,0.0003396121,0.0001712475,0.0004102375,0.0008155004,0.2377632,0.00488791,0.6328833,0.004941009,0.1110441],"study_design_scores_gemma":[0.00001158713,0.00006937009,0.0007093782,0.00009200618,0.00002141558,0.000163736,0.00006128618,0.5192896,0.003082486,0.4737996,0.002671273,0.00002842184],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.0279661,0.0009822184,0.9655748,0.001457229,0.00008570816,0.00004264101,0.00007378071,0.0003808331,0.00343672],"genre_scores_gemma":[0.7461703,0.001697371,0.2448345,0.001419045,0.0004442659,0.0002671297,0.0003941685,0.0007206005,0.004052576],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01370222,"threshold_uncertainty_score":0.07246518,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1198908800877382,"score_gpt":0.2356930585911815,"score_spread":0.1158021785034433,"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."}}