{"id":"W4385680786","doi":"10.48550/arxiv.2308.02594","title":"SMARLA: A Safety Monitoring Approach for Deep Reinforcement Learning Agents","year":2023,"lang":"en","type":"preprint","venue":"arXiv (Cornell University)","topic":"Adversarial Robustness in Machine Learning","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Natural Sciences and Engineering Research Council of Canada; Science Foundation Ireland","keywords":"Reinforcement learning; Reinforcement; Computer science; Safety monitoring; Artificial intelligence; Psychology; Social psychology","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0008284869,0.0004970118,0.0005140729,0.0004095323,0.0006562937,0.0002221623,0.002720781,0.0004034573,0.00001060421],"category_scores_gemma":[0.0002948655,0.0006169241,0.0003851289,0.0007462839,0.00007416365,0.0004711351,0.005121951,0.001394722,0.00005239776],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006453285,"about_ca_system_score_gemma":0.0001678538,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001447906,"about_ca_topic_score_gemma":0.000003460712,"domain_scores_codex":[0.9967502,0.0002306301,0.0003890988,0.001677816,0.0002245499,0.0007277373],"domain_scores_gemma":[0.9975237,0.0003157663,0.0005174477,0.001231609,0.0001995823,0.0002119212],"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":[0.00005297037,0.00002745676,0.005414806,0.000194737,0.0001565761,0.00007145682,0.0005221731,0.9675199,0.000004151498,0.02504591,0.00004638973,0.0009435142],"study_design_scores_gemma":[0.000790463,0.00005814762,0.001024485,0.0001168043,0.00008597528,0.000001818589,0.000269793,0.9926977,0.00001947578,0.003830537,0.0005129139,0.0005919575],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.004074509,0.00002784111,0.9902264,0.00005549764,0.001697735,0.0007800522,0.000002097446,0.0009357336,0.002200105],"genre_scores_gemma":[0.9470459,0.0001308908,0.04565805,0.00002029663,0.0003682903,0.0000118517,0.00007428106,0.00007033685,0.006620151],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9445684,"threshold_uncertainty_score":0.9996282,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1158806875982378,"score_gpt":0.231981940174605,"score_spread":0.1161012525763672,"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."}}