{"id":"W4417035278","doi":"10.48550/arxiv.2512.04006","title":"Diagonalizing the Softmax: Hadamard Initialization for Tractable Cross-Entropy Dynamics","year":2025,"lang":"","type":"preprint","venue":"ArXiv.org","topic":"Stochastic Gradient Optimization Techniques","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Centre International de Mathématiques et Informatique de Toulouse; Natural Sciences and Engineering Research Council of Canada","keywords":"Initialization; Artificial neural network; Softmax function; Hadamard transform; Convex function; Convex optimization; Spurious relationship; Function (biology); Regular polygon","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","sts","scholarly_communication"],"consensus_categories":[],"category_scores_codex":[0.001642975,0.0009819355,0.0009125023,0.0004622495,0.001710819,0.001790436,0.003993063,0.0008023693,0.0000972435],"category_scores_gemma":[0.001889931,0.0009136777,0.0006512692,0.001267663,0.0005474375,0.001007659,0.002569561,0.00105812,0.00003597459],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000871402,"about_ca_system_score_gemma":0.001056847,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001378475,"about_ca_topic_score_gemma":0.00002978039,"domain_scores_codex":[0.9937994,0.00031823,0.001770686,0.002081098,0.000847245,0.001183316],"domain_scores_gemma":[0.9931518,0.001753484,0.001385031,0.002091342,0.001395937,0.0002224318],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0002463151,0.00081661,0.159624,0.00135445,0.0007048001,0.00001377139,0.003679222,0.1733374,0.0001047461,0.6349862,0.003631469,0.02150098],"study_design_scores_gemma":[0.000948647,0.0001678619,0.01395465,0.000744965,0.0002531026,0.00001387586,0.00008250931,0.9430138,0.002212022,0.03237962,0.00517552,0.001053363],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.008670515,0.0004412826,0.9767361,0.003074342,0.004939299,0.003795857,0.0002997215,0.0007281644,0.001314751],"genre_scores_gemma":[0.8421898,0.00103898,0.1399595,0.003803047,0.001140115,0.003317185,0.0009635863,0.0001890423,0.007398737],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.8367766,"threshold_uncertainty_score":0.9995888,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05876001954424071,"score_gpt":0.3319350698524984,"score_spread":0.2731750503082577,"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."}}