{"id":"W3139073295","doi":"10.48550/arxiv.2103.07945","title":"Learning One Representation to Optimize All Rewards","year":2021,"lang":"en","type":"preprint","venue":"arXiv (Cornell University)","topic":"Reinforcement Learning in Robotics","field":"Computer Science","cited_by":5,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Montréal","funders":"","keywords":"Computer science; Representation (politics); Markov decision process; Artificial intelligence; Reinforcement learning; A priori and a posteriori; Range (aeronautics); Machine learning; Markov process; 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":[0.0003766225,0.0003183154,0.0004019037,0.0003340939,0.0001728173,0.0004684431,0.001922642,0.0002853564,0.00008235477],"category_scores_gemma":[0.000289331,0.0004373427,0.0002271272,0.0008789368,0.00004650778,0.0005165689,0.004456491,0.001036148,0.0001816092],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003528249,"about_ca_system_score_gemma":0.0002302582,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002000585,"about_ca_topic_score_gemma":0.00001013525,"domain_scores_codex":[0.9972544,0.0003211893,0.0002944985,0.001427375,0.0002466041,0.000455945],"domain_scores_gemma":[0.9973011,0.0001501135,0.0003266416,0.001629487,0.0003290166,0.0002636487],"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":[0.00001451635,0.00003094896,0.001087305,0.00003918617,0.0001222552,0.0002469119,0.001006214,0.9899661,0.00008932516,0.006825799,0.0001991015,0.0003723718],"study_design_scores_gemma":[0.0003420612,0.000102036,0.0008724043,0.0001465259,0.00007120627,0.000003876019,0.0002534992,0.9960559,0.0003164761,0.0004833007,0.0008534401,0.0004992512],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.0543158,0.00001734397,0.9376177,0.0003250478,0.0006632416,0.0003184426,8.135451e-7,0.000422308,0.006319299],"genre_scores_gemma":[0.9346163,0.0001667716,0.05431176,0.0002753396,0.00008666405,0.000001722342,0.00005308058,0.00002898085,0.01045939],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.883306,"threshold_uncertainty_score":0.9998078,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1225321745249543,"score_gpt":0.2230164994008293,"score_spread":0.1004843248758749,"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."}}