{"id":"W7160043136","doi":"10.1109/iccv51701.2025.00137","title":"Controlling Multimodal Llms Via Reward-Guided Decoding","year":2025,"lang":"","type":"article","venue":"","topic":"Speech and dialogue systems","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"Mila - Quebec Artificial Intelligence Institute","funders":"","keywords":"Decoding methods; Encoding (memory); Key (lock); Action (physics)","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":[0.0006317764,0.0007986972,0.0006317786,0.0002638036,0.000423265,0.001327155,0.0008451242,0.0007929338,0.005686899],"category_scores_gemma":[0.005557652,0.000271323,0.0001968208,0.000181519,0.0004778484,0.001125039,0.001482341,0.001264026,0.001789095],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003714346,"about_ca_system_score_gemma":0.0006473915,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001390469,"about_ca_topic_score_gemma":0.00222001,"domain_scores_codex":[0.9995198,0.0001418923,0.00002879687,0.0001077797,0.0001232769,0.00007829503],"domain_scores_gemma":[0.9983512,0.001088459,0.0001039499,0.0001164607,0.0002376205,0.0001022681],"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":[0.001864684,0.0003234659,0.002069281,0.0002966923,0.00006254911,0.0005101712,0.0007729355,0.153264,0.3784446,0.02466942,0.006586831,0.4311354],"study_design_scores_gemma":[0.00003883992,0.00009377035,0.0002969731,0.00001722302,0.00001376432,0.0000668095,0.00005094148,0.9444771,0.04419096,0.00877551,0.001952397,0.0000256798],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.07248868,0.0002883168,0.9095328,0.0004329261,0.0002721888,0.00008148427,0.0001063604,0.006842426,0.009954835],"genre_scores_gemma":[0.9284251,0.00007436734,0.06557943,0.0001300339,0.00005756283,0.00006530354,0.00006289902,0.0006295067,0.004975675],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.005686899,"threshold_uncertainty_score":0.01902461,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02146356907330903,"score_gpt":0.2809587149882836,"score_spread":0.2594951459149746,"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."}}