{"id":"W3176864543","doi":"10.65109/rgis8995","title":"Transferable Environment Poisoning: Training-time Attack on Reinforcement Learning","year":2021,"lang":"en","type":"article","venue":"","topic":"Adversarial Robustness in Machine Learning","field":"Computer Science","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"","keywords":"Reinforcement; Training (meteorology); Reinforcement learning; Computer science; Computer security; Psychology; Artificial intelligence; Social psychology; Geography; Meteorology","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":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0004303156,0.0002351132,0.0002594495,0.00007854088,0.0003426958,0.0001810878,0.0005626914,0.00008453071,0.002017276],"category_scores_gemma":[0.0001206735,0.000230285,0.0001200724,0.00024348,0.00003474781,0.0003701891,0.0002342186,0.000528608,0.0009190069],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001279928,"about_ca_system_score_gemma":0.0001004371,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000009729284,"about_ca_topic_score_gemma":6.598728e-7,"domain_scores_codex":[0.9978131,0.0001924814,0.0003136147,0.0006341076,0.0005437014,0.0005029777],"domain_scores_gemma":[0.9989893,0.0002088644,0.00006993294,0.0005583127,0.00002937703,0.0001442318],"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":[0.000007744766,0.00003681047,0.00008471179,0.000005906143,0.00002983578,0.00007427134,0.001722525,0.957492,0.00122858,0.01962105,0.0003390297,0.01935757],"study_design_scores_gemma":[0.0007872637,0.0002733228,0.0001608546,0.00004124895,0.00001086543,0.00003068613,0.0002503938,0.9282169,0.004868703,0.0002414638,0.064707,0.0004113287],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.004389013,0.0000305749,0.9020444,0.002522199,0.0001879674,0.0001094086,1.470651e-7,0.0003167417,0.09039959],"genre_scores_gemma":[0.8663451,0.00002648741,0.08341561,0.001242048,0.0001281571,0.00001732208,0.00001444895,0.00003304659,0.04877776],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.8619561,"threshold_uncertainty_score":0.9998589,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03446290798627397,"score_gpt":0.2597096128112343,"score_spread":0.2252467048249604,"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."}}