{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.001504018,0.0005011923,0.0008292377,0.0004749631,0.00057517,0.0009577157,0.001666209,0.0003421785,0.0001840761],"category_scores_gemma":[0.0004791648,0.0004747813,0.0004090441,0.00127869,0.0001195967,0.0006793117,0.0006212897,0.0003509759,0.0007714677],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002304851,"about_ca_system_score_gemma":0.0005690334,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001840033,"about_ca_topic_score_gemma":0.0001746895,"domain_scores_codex":[0.9954491,0.000301612,0.001299847,0.001236368,0.0005500506,0.001162985],"domain_scores_gemma":[0.9970738,0.0005699399,0.0002728057,0.001335623,0.0003852711,0.000362587],"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":[0.0004263252,0.0007887693,0.02101661,0.0006516572,0.001081389,0.0003688358,0.003379078,0.00575112,0.03596821,0.2265091,0.04787036,0.6561885],"study_design_scores_gemma":[0.003882434,0.000106043,0.000955824,0.000332875,0.00005657364,0.00003379969,0.00010886,0.9705942,0.01310653,0.005033864,0.00517688,0.0006121075],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.006587749,0.00156536,0.8920923,0.001781616,0.009724198,0.0008140703,0.00000403805,0.0003665534,0.08706409],"genre_scores_gemma":[0.9460421,0.00004621941,0.04311439,0.001784972,0.0004980771,0.00003333485,0.000003199865,0.00001880297,0.00845889],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9648431,"threshold_uncertainty_score":0.9997704,"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."}}