{"id":"W4416035046","doi":"10.18653/v1/2025.findings-emnlp.387","title":"AdaptMerge: Inference Time Adaptive Visual and Language-Guided Token Merging for Efficient Large Multimodal Models","year":2025,"lang":"","type":"article","venue":"","topic":"Multimodal Machine Learning Applications","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Saskatchewan","funders":"Natural Sciences and Engineering Research Council of Canada; DeepMind","keywords":"Inference; Security token; Feature (linguistics); Visualization; Pattern recognition (psychology)","routes":{"ca_aff":true,"ca_fund":true,"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":[0.001270639,0.001755397,0.001414184,0.0008768088,0.0007866222,0.001418591,0.003833246,0.001469682,0.008256493],"category_scores_gemma":[0.004785937,0.0008405385,0.001451222,0.0008881798,0.0009624937,0.003378362,0.003200204,0.002694232,0.003058389],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001605384,"about_ca_system_score_gemma":0.002174964,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01179247,"about_ca_topic_score_gemma":0.02399725,"domain_scores_codex":[0.999331,0.0001700431,0.00003308713,0.000232723,0.0001384011,0.00009470745],"domain_scores_gemma":[0.9991062,0.0004325839,0.00004469894,0.0002293614,0.0001216929,0.00006544884],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0007550836,0.0002200807,0.00137309,0.0003252492,0.0002404138,0.000293973,0.0002825948,0.339612,0.01637983,0.0160864,0.02600586,0.5984254],"study_design_scores_gemma":[0.00003468794,0.00004411033,0.0001173159,0.00001050175,0.0000220281,0.00005124688,0.00004285848,0.977779,0.005444339,0.01344431,0.002992293,0.00001726067],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.01554338,0.0007117488,0.9599031,0.0003877513,0.0001602484,0.0001248827,0.0005407591,0.02068543,0.001942764],"genre_scores_gemma":[0.3178131,0.0003871428,0.6669927,0.0007950941,0.0001308164,0.0004272152,0.002802275,0.002831649,0.00782004],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.01179247,"threshold_uncertainty_score":0.02762073,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0188835584601864,"score_gpt":0.3386773881967409,"score_spread":0.3197938297365545,"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."}}