{"id":"W4386126338","doi":"10.1007/s10994-023-06368-z","title":"Cautious policy programming: exploiting KL regularization for monotonic policy improvement in reinforcement learning","year":2023,"lang":"en","type":"article","venue":"Machine Learning","topic":"Reinforcement Learning in Robotics","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"Japan Society for the Promotion of Science","keywords":"Reinforcement learning; Monotonic function; Computer science; Mathematical optimization; Regularization (linguistics); Bellman equation; Entropy (arrow of time); Maximization; Upper and lower bounds; Exploit; Artificial intelligence; Mathematics","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":[0.001620794,0.0003884798,0.0003762859,0.001513009,0.0007072092,0.0004812061,0.0008275995,0.0001484682,0.000007689317],"category_scores_gemma":[0.002085635,0.0004244015,0.0001353951,0.002976338,0.00004557974,0.0005990695,0.000769856,0.0007320837,0.00006062848],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007082804,"about_ca_system_score_gemma":0.0003482175,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007558073,"about_ca_topic_score_gemma":0.00002552704,"domain_scores_codex":[0.9962991,0.0001881322,0.000836871,0.0007455106,0.0006409673,0.001289349],"domain_scores_gemma":[0.9984078,0.0002549233,0.0005125187,0.0005258266,0.0001341569,0.0001647441],"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":[0.00001270683,0.00001614887,0.004871021,0.0000923164,0.00002527126,0.000009722112,0.002248586,0.8970615,0.001223485,0.02491965,0.00002075439,0.06949887],"study_design_scores_gemma":[0.001190706,0.0006953932,0.0009546346,0.00009464132,0.000008343748,0.000005834457,0.0002380562,0.9772773,0.0005743291,0.0006633642,0.01787413,0.0004233024],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.01104847,0.00004146834,0.9820335,0.002314174,0.000189188,0.001270729,4.994071e-7,0.001468888,0.001633077],"genre_scores_gemma":[0.9734272,0.00008138509,0.01347566,0.0002614091,0.000341455,0.0004352007,0.0001714327,0.0000896411,0.01171655],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9685578,"threshold_uncertainty_score":0.9998208,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01774184896903942,"score_gpt":0.2845911047478464,"score_spread":0.266849255778807,"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."}}