{"id":"W6979256277","doi":"","title":"SGD as Free Energy Minimization: A Thermodynamic View on Neural Network Training","year":2025,"lang":"en","type":"article","venue":"ArXiv.org","topic":"Stochastic Gradient Optimization Techniques","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Institut de Valorisation des Données; Canada First Research Excellence Fund","keywords":"Maxima and minima; Artificial neural network; Monotonic function; Stochastic gradient descent; Gradient descent; Energy (signal processing); Entropy (arrow of time); Function (biology)","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":[],"consensus_categories":[],"category_scores_codex":[0.0001313301,0.0002040887,0.000216548,0.000156504,0.0001962606,0.00008990384,0.001160473,0.00008939778,0.00004007389],"category_scores_gemma":[0.0001345543,0.0001968014,0.00008112829,0.001006851,0.00005618087,0.0002256088,0.0002925256,0.0001290221,0.0000194544],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000588145,"about_ca_system_score_gemma":0.00009150262,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001945809,"about_ca_topic_score_gemma":0.000006808085,"domain_scores_codex":[0.998557,0.00009218002,0.000300413,0.0004901059,0.0002091833,0.0003510927],"domain_scores_gemma":[0.9987873,0.0001916563,0.0001134013,0.0007656837,0.00006996362,0.00007200351],"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":[0.00001554671,0.00007145278,0.002913735,0.00001315714,0.00004753299,0.00001573871,0.0005008265,0.03515611,0.00006111553,0.9159325,0.00517043,0.04010189],"study_design_scores_gemma":[0.0006106382,0.0002214187,0.01243067,0.0002495043,0.00002385334,0.00001567857,0.00003105956,0.9310589,0.0003339253,0.04948448,0.005116626,0.000423242],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.01737615,0.0002979577,0.9661595,0.002730259,0.0006984038,0.0001793918,0.000001195502,0.0007348017,0.01182236],"genre_scores_gemma":[0.9649858,0.00004533167,0.02727311,0.006439613,0.0001282045,0.00008563694,0.000008079634,0.00002097593,0.001013207],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9476097,"threshold_uncertainty_score":0.8025326,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0284876120754896,"score_gpt":0.255307076511693,"score_spread":0.2268194644362034,"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."}}