{"id":"W4416036669","doi":"10.18653/v1/2025.emnlp-main.368","title":"LEO-MINI: An Efficient Multimodal Large Language Model using Conditional Token Reduction and Mixture of Multi-Modal Experts","year":2025,"lang":"","type":"article","venue":"","topic":"Topic Modeling","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada; Government of Canada","keywords":"Reduction (mathematics); Language model; Security token; Data modeling; Natural language","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":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0004497184,0.0003406248,0.0004037418,0.0003140747,0.000338925,0.0001570587,0.0004870768,0.000274561,0.00005908137],"category_scores_gemma":[0.00004449064,0.0003438178,0.0001150553,0.0003204567,0.0001632253,0.0005726123,0.0005313146,0.0002430889,0.000001304589],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001260368,"about_ca_system_score_gemma":0.0003622925,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003136887,"about_ca_topic_score_gemma":0.00003765586,"domain_scores_codex":[0.9972041,0.0001496694,0.0006661852,0.001001509,0.0004584831,0.0005200872],"domain_scores_gemma":[0.9986045,0.00004539045,0.000190929,0.0007315821,0.0002451467,0.0001824698],"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":[0.0000480822,0.000960548,0.0002361028,0.0001290706,0.00006020377,0.00001174107,0.01886347,0.8867077,0.06358699,0.02071873,0.00006407775,0.008613235],"study_design_scores_gemma":[0.00169746,0.00004726934,0.0002595647,0.0001211606,0.00003685555,0.00003143529,0.002287161,0.9838864,0.01101973,0.0002897408,0.0000118577,0.0003113871],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.4747596,0.0004550604,0.523762,0.0001382799,0.0004247179,0.0002459875,0.00004285826,0.00004758123,0.000123865],"genre_scores_gemma":[0.7815653,0.00000821722,0.2176135,0.0001562474,0.00009854941,0.000007974404,0.00001902066,0.00001257942,0.0005185685],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.3068057,"threshold_uncertainty_score":0.9999014,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02742399061934513,"score_gpt":0.3145264251647307,"score_spread":0.2871024345453855,"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."}}