{"id":"W4402811680","doi":"10.1109/csr61664.2024.10679438","title":"When Rewards Deceive: Counteracting Reward Poisoning on Online Deep Reinforcement Learning","year":2024,"lang":"en","type":"article","venue":"","topic":"Psychology of Moral and Emotional Judgment","field":"Neuroscience","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université du Québec en Outaouais","funders":"","keywords":"Reinforcement learning; Reinforcement; Computer science; Cognitive psychology; Artificial intelligence; Human–computer interaction; Psychology; Social psychology","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":[0.001968776,0.0008699562,0.0005702214,0.0002960722,0.0003638396,0.0007054348,0.001153296,0.0009857621,0.001515406],"category_scores_gemma":[0.009945344,0.0002976203,0.0002142014,0.0001086766,0.001307317,0.001353482,0.001900602,0.001733556,0.0002920191],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006541138,"about_ca_system_score_gemma":0.001043132,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001386817,"about_ca_topic_score_gemma":0.001939643,"domain_scores_codex":[0.9991721,0.0003128142,0.00003170869,0.0001517327,0.0001779211,0.0001537559],"domain_scores_gemma":[0.9967412,0.00155257,0.0007016344,0.0003660881,0.0003590659,0.0002794209],"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":[0.0005862167,0.0004116582,0.007234687,0.0001513032,0.0001287241,0.0004039616,0.0003338821,0.8172329,0.02202392,0.02940071,0.003004332,0.1190878],"study_design_scores_gemma":[0.00001987549,0.0001132972,0.0003657597,0.00001171475,0.000009911655,0.00003919724,0.00002074433,0.9867993,0.001906774,0.0103302,0.0003734394,0.00000972078],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2348366,0.0003565587,0.7549598,0.001215997,0.0001154331,0.0001294635,0.00004449061,0.001430527,0.006911168],"genre_scores_gemma":[0.9820702,0.00004448661,0.01667106,0.0001684292,0.00001180262,0.00002908073,0.00001062444,0.00002321932,0.0009711387],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.001968776,"threshold_uncertainty_score":0.01041198,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08086813724385629,"score_gpt":0.3253869984092584,"score_spread":0.2445188611654022,"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."}}