{"id":"W2995434136","doi":"10.48550/arxiv.1905.12558","title":"Limitations of the Empirical Fisher Approximation for Natural Gradient\\n Descent","year":2019,"lang":"","type":"preprint","venue":"arXiv (Cornell University)","topic":"Stochastic Gradient Optimization Techniques","field":"Computer Science","cited_by":30,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Fisher information; Hessian matrix; Fisher kernel; Heuristics; Mathematics; Applied mathematics; Econometrics; Mathematical economics; Mathematical optimization; Statistics; Computer science; Artificial intelligence; Kernel method","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":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0004195444,0.0004788336,0.000519792,0.0003823271,0.0003597344,0.0001330623,0.002809469,0.0003811918,0.00001347173],"category_scores_gemma":[0.0005591364,0.00045375,0.0006982255,0.001470693,0.0003923696,0.0005394426,0.001684475,0.0005882217,0.00001696119],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004926684,"about_ca_system_score_gemma":0.0004233898,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000226636,"about_ca_topic_score_gemma":0.00001457673,"domain_scores_codex":[0.9969797,0.0002661499,0.00060647,0.001389274,0.0002514834,0.0005069967],"domain_scores_gemma":[0.9950235,0.0008445903,0.001045729,0.00192306,0.001025203,0.0001378795],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000101143,0.0004695571,0.004628053,0.0002884351,0.0001929311,0.000002608863,0.002245918,0.6177351,0.00004979255,0.3710074,0.001620823,0.001658253],"study_design_scores_gemma":[0.000770342,0.0001497717,0.004289957,0.0002379312,0.0002099388,0.000003667512,0.0001617473,0.9582177,0.0005732715,0.03448184,0.0004477348,0.000456147],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.04539831,0.00004731787,0.9476411,0.0005301828,0.002581779,0.002875885,0.0000430706,0.0001482084,0.0007340957],"genre_scores_gemma":[0.9680316,0.0001060523,0.02976692,0.0001554862,0.00005056897,0.00001280301,0.00003702285,0.00003217711,0.001807396],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9226332,"threshold_uncertainty_score":0.9997914,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2954199039827467,"score_gpt":0.2289787568300307,"score_spread":0.06644114715271598,"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."}}