{"id":"W7163682316","doi":"10.13182/t130-44811","title":"Developing a Machine Learning Benchmark Using Real-Time Data from the PUR-1 Reactor for Nuclear Applications","year":2024,"lang":"","type":"article","venue":"Transactions of the American Nuclear Society","topic":"Nuclear reactor physics and engineering","field":"Engineering","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"McMaster University; CNIB Foundation; Canadian Nuclear Laboratories","funders":"","keywords":"Benchmark (surveying); Nuclear reactor; Support vector machine; Experimental data","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":[0.0003487943,0.0004758495,0.000586622,0.00004776684,0.0009091258,0.0003188422,0.001737097,0.000129938,0.0002843237],"category_scores_gemma":[0.00002047629,0.0003894173,0.0007537745,0.001248962,0.0004678133,0.0004472811,0.0002000084,0.001051738,0.00006255754],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004624058,"about_ca_system_score_gemma":0.0001458307,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003410003,"about_ca_topic_score_gemma":0.00002099403,"domain_scores_codex":[0.9977475,0.00007659239,0.0005965888,0.0006703158,0.0003539701,0.0005550521],"domain_scores_gemma":[0.997337,0.0005920964,0.0002377718,0.001638313,0.00007833927,0.0001164667],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"bench_or_experimental","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0001650227,0.0003638123,0.00005116297,0.001997024,0.01027022,0.000002594061,0.01801785,0.1256026,0.6997312,0.009609207,0.01055975,0.1236296],"study_design_scores_gemma":[0.0001479448,0.00003845912,0.0001363225,0.00030777,0.0006852513,0.000006450054,0.001897172,0.7586148,0.0001066475,0.0001153603,0.2375431,0.0004006881],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.4041511,0.004422039,0.5607836,0.006041707,0.002851058,0.004680828,0.01197193,0.003033424,0.002064322],"genre_scores_gemma":[0.9528539,0.004146101,0.04144862,0.0001054244,0.000589258,0.00002333186,0.0001453291,0.0006039587,0.00008411583],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.6996245,"threshold_uncertainty_score":0.9998558,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02612088275324759,"score_gpt":0.2532038301264672,"score_spread":0.2270829473732196,"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."}}