{"id":"W4415445607","doi":"10.1016/j.neucom.2025.131861","title":"Reducing hallucination in MLLMs for action recognition with adaptive masking and dual prompting","year":2025,"lang":"en","type":"article","venue":"Neurocomputing","topic":"Medical Imaging Techniques and Applications","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":false,"ca_institutions":"Artificial Intelligence in Medicine (Canada)","funders":"Natural Science Foundation of Guangdong Province; Beijing Normal University; National Natural Science Foundation of China","keywords":"Masking (illustration); Action (physics); Focus (optics); Dual (grammatical number); Backward masking; Visual masking; Filter (signal processing)","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":[0.0002372606,0.0000738955,0.0001193782,0.0001378896,0.0001077417,0.0000209348,0.00001908842,0.00003847134,0.000001359782],"category_scores_gemma":[0.0001246373,0.00006752219,0.00001414459,0.000216719,0.0000237792,0.00006120089,0.00002476759,0.0001801751,2.960546e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004372036,"about_ca_system_score_gemma":0.00003588362,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002987873,"about_ca_topic_score_gemma":0.000004541599,"domain_scores_codex":[0.9993425,0.00001925675,0.0001763677,0.0002496168,0.00007732332,0.0001349502],"domain_scores_gemma":[0.9996179,0.0001368283,0.00007365082,0.00006971781,0.00007166847,0.00003017212],"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":[0.0001688703,0.0001159461,0.002804023,0.0004005518,0.00001594242,0.0000101342,0.0003708341,0.0001774271,0.03017792,0.0007576273,0.0002158872,0.9647849],"study_design_scores_gemma":[0.002282283,0.0003799242,0.02790032,0.003113605,0.00009169584,0.0001172559,0.0003621866,0.9443995,0.01859446,0.001611887,0.0009610184,0.0001858008],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6907358,0.00001243624,0.3052099,0.002214389,0.00003041482,0.0008303331,6.50212e-7,0.0001263159,0.0008397812],"genre_scores_gemma":[0.8601,0.000004699503,0.1394069,0.0002683848,0.00007291323,0.00008740799,0.00001164965,0.000009578899,0.00003846742],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.964599,"threshold_uncertainty_score":0.2753475,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05718725923353744,"score_gpt":0.3444681618641812,"score_spread":0.2872809026306437,"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."}}