{"id":"W4405055855","doi":"10.1109/lra.2024.3512374","title":"Safety Filtering While Training: Improving the Performance and Sample Efficiency of Reinforcement Learning Agents","year":2024,"lang":"en","type":"article","venue":"IEEE Robotics and Automation Letters","topic":"Occupational Health and Safety Research","field":"Health Professions","cited_by":13,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Training (meteorology); Sample (material); Reinforcement learning; Reinforcement; Computer science; Artificial intelligence; Psychology; Chromatography; Social psychology; Geography; Chemistry","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":[],"consensus_categories":[],"category_scores_codex":[0.0008685847,0.00007977698,0.0001143785,0.00008413121,0.0007711107,0.00002462492,0.00006110449,0.00004438552,0.00004599036],"category_scores_gemma":[0.0001276763,0.00005817445,0.00002002245,0.0001446757,0.00005858499,0.0001282418,0.00004567562,0.0003319718,0.00000801847],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005389083,"about_ca_system_score_gemma":0.0001215528,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00008379698,"about_ca_topic_score_gemma":0.000003579992,"domain_scores_codex":[0.9988576,0.0001111583,0.0003785866,0.0001446467,0.0002323884,0.0002756148],"domain_scores_gemma":[0.9987086,0.0009972949,0.00009342436,0.0000868599,0.00004712295,0.00006668751],"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":[0.0001069438,0.00002402984,0.038893,0.007482022,0.00006471199,0.000004062694,0.0323051,0.7511612,0.01182431,0.004391509,0.001507646,0.1522355],"study_design_scores_gemma":[0.0002206791,0.00006891866,0.05403211,0.0003384652,0.000008495344,0.000001471941,0.0005944829,0.9435377,0.00004975401,0.00001482243,0.001063593,0.00006954305],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9064565,0.000100241,0.08614743,0.005819718,0.0005343104,0.0005557776,0.00000858207,0.00007106428,0.0003064083],"genre_scores_gemma":[0.9981514,0.0001457034,0.001015141,0.0004693081,0.0001057203,0.00002557743,0.00001232608,0.00001017046,0.00006464173],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1923765,"threshold_uncertainty_score":0.5930838,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07189234968498567,"score_gpt":0.3722205461360454,"score_spread":0.3003281964510597,"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."}}