{"id":"W4393160778","doi":"10.1609/aaai.v38i16.29742","title":"Unsupervised Layer-Wise Score Aggregation for Textual OOD Detection","year":2024,"lang":"en","type":"article","venue":"Proceedings of the AAAI Conference on Artificial Intelligence","topic":"Anomaly Detection Techniques and Applications","field":"Computer Science","cited_by":10,"is_retracted":false,"has_abstract":true,"ca_institutions":"Canadian Institute for Advanced Research; McGill University; Mila - Quebec Artificial Intelligence Institute","funders":"Grand Équipement National De Calcul Intensif","keywords":"Anomaly detection; Mahalanobis distance; Computer science; Oracle; Robustness (evolution); Encoder; Leverage (statistics); Layer (electronics); Artificial intelligence; Discriminator; Embedding; Benchmark (surveying); Data mining; Machine learning; Detector; Pattern recognition (psychology)","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":[0.00243143,0.001709074,0.001437286,0.002615026,0.0005143657,0.001338166,0.001951278,0.001033004,0.001415761],"category_scores_gemma":[0.008455924,0.0003495574,0.0008634072,0.001654139,0.0005820708,0.001802034,0.002087274,0.001391085,0.001036198],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009076505,"about_ca_system_score_gemma":0.0009776377,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005079435,"about_ca_topic_score_gemma":0.008657125,"domain_scores_codex":[0.9979454,0.0004018962,0.0001824111,0.0004870106,0.000699736,0.000283577],"domain_scores_gemma":[0.9953252,0.001526747,0.0006491009,0.0009520224,0.001295889,0.0002509969],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0008694812,0.0005035277,0.02047987,0.000206251,0.0002916622,0.0002107587,0.0001873424,0.06987698,0.03667963,0.001677761,0.007675004,0.8613417],"study_design_scores_gemma":[0.00002980671,0.0001712611,0.006079467,0.00001329633,0.00006207585,0.0001132006,0.00005990295,0.9570385,0.03109574,0.003560656,0.001739095,0.00003698314],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2484563,0.001376891,0.7240282,0.0003955016,0.0002070293,0.0002379178,0.001276945,0.02115118,0.002870004],"genre_scores_gemma":[0.753173,0.000188723,0.2394779,0.0001429233,0.000112388,0.0001167324,0.002879104,0.0004546858,0.003454504],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.005079435,"threshold_uncertainty_score":0.01285875,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08825742206946548,"score_gpt":0.3091717903066092,"score_spread":0.2209143682371438,"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."}}