{"id":"W2995509794","doi":"10.48550/arxiv.2002.06487","title":"Maxmin Q-learning: Controlling the Estimation Bias of Q-learning","year":2020,"lang":"en","type":"article","venue":"arXiv (Cornell University)","topic":"Reinforcement Learning in Robotics","field":"Computer Science","cited_by":38,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Convergence (economics); Benchmark (surveying); Computer science; Variance (accounting); Generalization error; Generalization; Artificial intelligence; Inductive bias; Machine learning; Q-learning; Algorithm; Value (mathematics); Mathematics; Stability (learning theory); Multi-task learning; Reinforcement learning","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":[0.0003491903,0.0001323102,0.0001802082,0.00007314641,0.0002376931,0.00007950456,0.0009065251,0.00005825939,0.00002751964],"category_scores_gemma":[0.0005287679,0.0001184972,0.0001005565,0.0008094757,0.00008543801,0.0004081578,0.000277817,0.0003946338,0.00009872568],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003601798,"about_ca_system_score_gemma":0.00005027074,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001485794,"about_ca_topic_score_gemma":5.585646e-7,"domain_scores_codex":[0.9988726,0.0002308456,0.0002075032,0.0003349006,0.0001285637,0.0002256152],"domain_scores_gemma":[0.9988005,0.0003814245,0.0003186989,0.0002939008,0.0001089685,0.00009651726],"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":[0.00001578436,0.000006071442,0.005913099,0.0000132053,0.0000248157,0.00001672465,0.0007239453,0.9316254,0.0001617559,0.06093642,0.00002946863,0.0005333453],"study_design_scores_gemma":[0.0004512741,0.0001659585,0.0005072656,0.00001621354,0.00002451254,0.000001501192,0.0002836241,0.9964485,0.0003120981,0.0002702047,0.001393641,0.0001251659],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.05948207,0.00001480723,0.9376166,0.00049605,0.00006834623,0.0001260504,1.589421e-7,0.0001806742,0.002015234],"genre_scores_gemma":[0.9963404,0.00002129969,0.002697897,0.000140163,0.00002730704,2.174906e-7,0.000002111242,0.000009737191,0.0007608751],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9368583,"threshold_uncertainty_score":0.4832177,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09264049702283787,"score_gpt":0.1833738516716606,"score_spread":0.09073335464882272,"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."}}