{"id":"W4413182473","doi":"10.1214/25-aap2187","title":"Spectral alignment of stochastic gradient descent for high-dimensional classification tasks","year":2025,"lang":"en","type":"article","venue":"The Annals of Applied Probability","topic":"Random Matrices and Applications","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"","keywords":"Stochastic gradient descent; Computer science; Artificial intelligence; Descent (aeronautics); Gradient descent; Pattern recognition (psychology); Physics; Artificial neural network","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":[0.003598884,0.0008199391,0.0009444742,0.0008003409,0.0006378998,0.001445749,0.001331016,0.001354789,0.00258395],"category_scores_gemma":[0.02344496,0.000636271,0.0005532053,0.0006995961,0.002414883,0.002909803,0.002199966,0.002003943,0.0006715882],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001561705,"about_ca_system_score_gemma":0.001223652,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002743911,"about_ca_topic_score_gemma":0.002376376,"domain_scores_codex":[0.9987183,0.0006264459,0.00005714572,0.0001964848,0.0002936053,0.000108098],"domain_scores_gemma":[0.9923316,0.004687187,0.001038352,0.000715786,0.0007460259,0.0004810659],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0001123369,0.00009175819,0.002932581,0.0001068024,0.00005911616,0.0001331543,0.0003103502,0.7558666,0.003825582,0.2117983,0.001445568,0.02331793],"study_design_scores_gemma":[0.000003113088,0.00001695847,0.0002751816,0.000006801587,0.000001549334,0.00001096361,0.00001075305,0.9629718,0.0002470408,0.0362828,0.0001673152,0.000005671856],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.07440017,0.0003249459,0.9211839,0.0005689978,0.00004414277,0.0000333608,0.00005802147,0.0002860578,0.003100524],"genre_scores_gemma":[0.8900304,0.0004232222,0.1043543,0.0002251269,0.00009490827,0.0001406472,0.0002206887,0.0002559907,0.004254686],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.003598884,"threshold_uncertainty_score":0.01903296,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1170952197230851,"score_gpt":0.3545861818715112,"score_spread":0.237490962148426,"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."}}