{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003219415,0.0008735629,0.001092984,0.0005584392,0.000456806,0.0009710325,0.001502228,0.001339398,0.001439414],"category_scores_gemma":[0.01134026,0.0004942639,0.000409727,0.0004470735,0.002212133,0.001488423,0.001872046,0.002664257,0.0002334457],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009406584,"about_ca_system_score_gemma":0.001623935,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001874886,"about_ca_topic_score_gemma":0.001421709,"domain_scores_codex":[0.9988149,0.0005271801,0.00004750833,0.0001998759,0.0002935624,0.0001168713],"domain_scores_gemma":[0.9956293,0.002981045,0.0004320747,0.0003078386,0.0004358486,0.00021398],"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":[0.000109831,0.00008610002,0.0006785615,0.00006022368,0.00003659529,0.000063819,0.00009729919,0.9074284,0.003172509,0.04271054,0.000880099,0.04467608],"study_design_scores_gemma":[0.000007382268,0.00002247797,0.00002189679,0.000004314803,0.000001578927,0.000005402389,0.000002047537,0.9936664,0.0003670987,0.00577041,0.0001280312,0.000002910623],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.01280132,0.00008983615,0.9854733,0.0002130444,0.00002118954,0.00003358757,0.00000735429,0.0001519296,0.001208451],"genre_scores_gemma":[0.7595494,0.0001082956,0.2376505,0.0003380696,0.00005217598,0.0001878915,0.00003342866,0.0001320263,0.001948258],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.003219415,"threshold_uncertainty_score":0.01702607,"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."}}