{"id":"W4391903608","doi":"10.24251/hicss.2023.799","title":"Safe Reinforcement Learning via Observation Shielding","year":2023,"lang":"en","type":"article","venue":"Proceedings of the ... Annual Hawaii International Conference on System Sciences/Proceedings of the Annual Hawaii International Conference on System Sciences","topic":"Adversarial Robustness in Machine Learning","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université Laval","funders":"National Science Foundation","keywords":"Robustness (evolution); Computer science; Adversarial system; Agile software development; SAFER; Artificial intelligence; Machine learning; Reinforcement learning; Deep neural networks; Artificial neural network; Computer security","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","sts","scholarly_communication","open_science"],"consensus_categories":[],"category_scores_codex":[0.006482722,0.0008085742,0.0008908998,0.001336486,0.002024064,0.001525696,0.01496419,0.0003084611,0.00006850388],"category_scores_gemma":[0.000389934,0.0005592675,0.000532738,0.003926868,0.001536549,0.004692068,0.00327281,0.00103868,0.0001091695],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007265199,"about_ca_system_score_gemma":0.0005671366,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003252324,"about_ca_topic_score_gemma":0.000008702759,"domain_scores_codex":[0.988614,0.0001049331,0.002018032,0.001835353,0.006352232,0.001075467],"domain_scores_gemma":[0.9894305,0.0003802449,0.003606273,0.0004153372,0.0059266,0.0002410219],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"qualitative","study_design_scores_codex":[0.00009112687,0.00006325894,0.00887637,0.0001985612,0.00009521941,0.000001535499,0.00282733,0.00921315,0.001932283,0.9748735,0.0007809026,0.001046771],"study_design_scores_gemma":[0.001500994,0.001454733,0.01012237,0.00815771,0.00007669631,0.000177293,0.8395562,0.1253742,0.00324986,0.007379363,0.001419003,0.001531548],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.04786221,0.00001260632,0.002237431,0.009244497,0.005856317,0.001043231,0.00007202606,0.0005103851,0.9331613],"genre_scores_gemma":[0.9940469,0.00002857295,0.00170082,0.0001492723,0.0004557283,0.0001160914,0.000005724411,0.00003666145,0.003460263],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9674941,"threshold_uncertainty_score":0.9996859,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05429878908399725,"score_gpt":0.3038269335608544,"score_spread":0.2495281444768571,"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."}}