{"id":"W4414690632","doi":"10.1109/iccv51701.2025.01654","title":"ViLU: Learning Vision-Language Uncertainties for Failure Prediction","year":2025,"lang":"en","type":"preprint","venue":"","topic":"Infrastructure Maintenance and Monitoring","field":"Engineering","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"École de Technologie Supérieure","funders":"Grand Équipement National De Calcul Intensif; Agence Nationale de la Recherche","keywords":"Representation (politics); Uncertainty quantification; Classifier (UML); Binary number; Binary classification; Code (set theory); Bounding overwatch","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":[0.002309132,0.00223912,0.001569444,0.001809682,0.0005712245,0.001936679,0.003531663,0.001970342,0.003127199],"category_scores_gemma":[0.01035272,0.0008164715,0.00155408,0.001033398,0.001317021,0.004604892,0.003491964,0.003241072,0.001155842],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001867746,"about_ca_system_score_gemma":0.001801102,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007730478,"about_ca_topic_score_gemma":0.00953582,"domain_scores_codex":[0.9985785,0.0003377918,0.00007115599,0.0004843741,0.0003728578,0.0001553002],"domain_scores_gemma":[0.9971399,0.001549698,0.0003645789,0.000386937,0.0004161959,0.0001427766],"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":[0.0003089006,0.0002074775,0.003123951,0.0003449088,0.0001940783,0.0001952106,0.0003073605,0.6187295,0.006236749,0.02156361,0.01721604,0.3315723],"study_design_scores_gemma":[0.000005642286,0.00002708339,0.0001404133,0.00001838064,0.00000909557,0.00002144844,0.00001774496,0.9813489,0.001325222,0.01635579,0.0007169875,0.00001316083],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.008393969,0.0005146322,0.9827744,0.0003719844,0.00006872972,0.00006532163,0.0005586866,0.00642059,0.0008316705],"genre_scores_gemma":[0.6084827,0.0005260078,0.3808899,0.0008722435,0.0002530974,0.0003981318,0.003417038,0.001141393,0.004019585],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.007730478,"threshold_uncertainty_score":0.01537097,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.004061454110672231,"score_gpt":0.2381710535032332,"score_spread":0.234109599392561,"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."}}