{"id":"W4283797949","doi":"10.1609/aaai.v36i9.21233","title":"Stochastic Goal Recognition Design Problems with Suboptimal Agents","year":2022,"lang":"en","type":"article","venue":"Proceedings of the AAAI Conference on Artificial Intelligence","topic":"Reinforcement Learning in Robotics","field":"Computer Science","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"National Science Foundation","keywords":"Ambiguity; Computer science; Generalization; Observability; Action (physics); Benchmark (surveying); Artificial intelligence; Range (aeronautics); Mathematical optimization; Observer (physics); Machine learning; Mathematics; Engineering","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":[],"consensus_categories":[],"category_scores_codex":[0.000742035,0.0002314203,0.0002128686,0.0001596158,0.0005079584,0.0002598929,0.002148938,0.00003668233,0.0001845926],"category_scores_gemma":[0.0001928608,0.0001797779,0.00007399151,0.0008563429,0.0001942054,0.0004267856,0.0006616567,0.0005045254,0.00007924158],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001110372,"about_ca_system_score_gemma":0.0001540663,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001837774,"about_ca_topic_score_gemma":0.000001032391,"domain_scores_codex":[0.9976083,0.00004716043,0.0004702571,0.0005024269,0.0009663124,0.0004055316],"domain_scores_gemma":[0.9985246,0.0001068414,0.0005240828,0.0002896266,0.0004727837,0.00008207066],"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":[0.0001902248,0.0002350781,0.00005894945,0.00003671703,0.00003357282,0.000001716385,0.002310263,0.8525912,0.005417504,0.1212229,0.0001980997,0.01770388],"study_design_scores_gemma":[0.0000731677,0.001337904,0.00004987054,0.0001356405,0.00002114665,0.00002573273,0.0005309642,0.9339495,0.03649581,0.02702903,0.00004485539,0.0003064034],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.0243535,0.000005536273,0.9703914,0.00102252,0.0003362336,0.0008476899,0.000003308421,0.0001298316,0.002909967],"genre_scores_gemma":[0.9849914,0.000003883459,0.01430617,0.0001671096,0.00002955289,0.0001366627,0.000001426794,0.00001812105,0.000345668],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9606379,"threshold_uncertainty_score":0.7331129,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1258906047509866,"score_gpt":0.2762074138993237,"score_spread":0.1503168091483371,"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."}}