{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004409973,0.001523501,0.001738017,0.0006875556,0.0006474764,0.001580899,0.001354184,0.002257133,0.002352843],"category_scores_gemma":[0.01344803,0.001000241,0.001444728,0.0004810039,0.002109108,0.001691689,0.002409604,0.002067229,0.0003005161],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002000957,"about_ca_system_score_gemma":0.001806132,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00356705,"about_ca_topic_score_gemma":0.002675291,"domain_scores_codex":[0.9967443,0.001274752,0.0002544877,0.0008549513,0.0005395367,0.000331913],"domain_scores_gemma":[0.9903702,0.006709182,0.001194062,0.0005818868,0.0007517188,0.0003929081],"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":[0.00007635644,0.00003786361,0.0006466106,0.00008505447,0.00004034606,0.00008578094,0.00007757426,0.9611997,0.0009142273,0.02511672,0.0003178577,0.01140204],"study_design_scores_gemma":[0.00002707615,0.00005664916,0.00009688965,0.00001236536,0.00001170795,0.00001956028,0.00001777657,0.9761681,0.0004748155,0.02274952,0.0003566661,0.000008837965],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.02878604,0.0001911179,0.9682109,0.000327018,0.00002718154,0.00009602345,0.00005120484,0.0001466407,0.002163789],"genre_scores_gemma":[0.7794319,0.0002097041,0.2159711,0.0002669977,0.0000390313,0.0003946522,0.0001809194,0.00007600863,0.003429731],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.004409973,"threshold_uncertainty_score":0.0233224,"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."}}