{"id":"W3158681970","doi":"10.48550/arxiv.2104.13844","title":"A Generalized Projected Bellman Error for Off-policy Value Estimation in Reinforcement Learning","year":2021,"lang":"en","type":"preprint","venue":"arXiv (Cornell University)","topic":"Reinforcement Learning in Robotics","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Reinforcement learning; Bellman equation; Temporal difference learning; Function approximation; Hyperparameter; Nonlinear system; Mean squared error; Artificial neural network; Mathematical optimization; Function (biology); Computer science; Approximation error; Linear approximation; Value (mathematics); Mathematics; Applied mathematics; Algorithm; Artificial intelligence; Machine learning; Statistics","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":[0.0005141936,0.0004321449,0.0005097638,0.0008221802,0.0002389627,0.000301125,0.001461502,0.0003647853,0.00001859029],"category_scores_gemma":[0.0004219103,0.0005435501,0.0002275075,0.001283344,0.00006593793,0.0005578226,0.00196636,0.0008190741,0.00001681221],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000859472,"about_ca_system_score_gemma":0.000950857,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005714744,"about_ca_topic_score_gemma":0.00004519788,"domain_scores_codex":[0.9971901,0.0002930335,0.0005036896,0.001193383,0.0002066457,0.0006131547],"domain_scores_gemma":[0.9977406,0.0001591883,0.0005811203,0.001091021,0.0002883451,0.0001396842],"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":[0.00003206348,0.00003541446,0.0004447217,0.0001737029,0.00007656009,0.00006306239,0.0009351832,0.9247121,0.00005787368,0.07288083,0.00004407617,0.0005444688],"study_design_scores_gemma":[0.001346836,0.0001482897,0.0004009516,0.0002441731,0.00005087051,0.000002987484,0.00009520162,0.9951712,0.0001587252,0.001360998,0.0004915494,0.0005281465],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.07484218,0.0000374751,0.9221404,0.0001556664,0.0003430855,0.001150898,0.000001296262,0.0002969621,0.001032039],"genre_scores_gemma":[0.9554626,0.0001543423,0.03814177,0.0001277475,0.00006963983,0.00001692158,0.0002033556,0.00003728954,0.005786334],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8839986,"threshold_uncertainty_score":0.9997016,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07410652224032353,"score_gpt":0.2396721545053905,"score_spread":0.165565632265067,"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."}}