{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0003192459,0.000163132,0.0001426286,0.0001582133,0.0002531157,0.0003830693,0.0009736412,0.00009648203,0.00002292981],"category_scores_gemma":[0.0001230395,0.0001266025,0.0001491431,0.0008236059,0.0001166736,0.0004429915,0.0001391888,0.0001962761,0.00005202064],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005502336,"about_ca_system_score_gemma":0.00007421836,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002201244,"about_ca_topic_score_gemma":0.00001259085,"domain_scores_codex":[0.9986635,0.000007536809,0.0003853318,0.0004671217,0.0002609241,0.00021563],"domain_scores_gemma":[0.9990152,0.00008371483,0.0001451244,0.0002355918,0.0004641673,0.00005617126],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.00001261707,0.00003359394,0.000005810921,0.00004015695,0.000006284255,5.612828e-8,0.0002182668,0.000009499976,0.07603496,0.5850952,0.00008750174,0.3384561],"study_design_scores_gemma":[0.0000109339,0.0001655643,0.0000212367,0.0001444337,0.000009614311,0.000003618191,0.0001246684,0.169412,0.6803495,0.148753,0.0008918974,0.0001136069],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.05639767,0.00004768458,0.9357616,0.003028278,0.0003462969,0.0008946517,0.00001075374,0.000493634,0.003019495],"genre_scores_gemma":[0.9942675,0.00003627676,0.004943215,0.00009307881,0.0001064963,0.0002450168,7.187667e-7,0.00001353,0.0002941168],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9378699,"threshold_uncertainty_score":0.5162699,"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."}}