{"id":"W4415367566","doi":"10.1109/isit63088.2025.11195330","title":"Leveraging Conditional Mutual Information to Improve Large Language Model Fine-Tuning for Classification","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":"","keywords":"Mutual information; Process (computing); Adaptability; Language model; Information theory; Deep learning","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":[0.003672605,0.00159612,0.001358252,0.001278808,0.0006705722,0.001806238,0.002383986,0.001336107,0.001712544],"category_scores_gemma":[0.01400895,0.0006108263,0.0014391,0.000917448,0.001050969,0.003814465,0.003082934,0.003588958,0.001122522],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001378013,"about_ca_system_score_gemma":0.002232432,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003949816,"about_ca_topic_score_gemma":0.00806179,"domain_scores_codex":[0.9983955,0.0006191789,0.0001089763,0.0003891021,0.000321859,0.0001654006],"domain_scores_gemma":[0.9962805,0.002282131,0.0002701412,0.0005564255,0.0004219512,0.0001887606],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.0002182829,0.000280501,0.003257631,0.000219687,0.0002599589,0.0001045757,0.0002561922,0.6110148,0.01849255,0.01960414,0.005543157,0.3407485],"study_design_scores_gemma":[0.000007873654,0.00002788122,0.0001181959,0.000007521574,0.00001136584,0.00001234678,0.00001384262,0.9870603,0.003115668,0.009113487,0.0005003353,0.00001109717],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.02635112,0.0005487793,0.9681952,0.0004117191,0.00005957293,0.00005702015,0.0001258494,0.002929712,0.001321132],"genre_scores_gemma":[0.6232759,0.0003637509,0.3705826,0.0007068968,0.0001517862,0.0003009675,0.001050363,0.0008962568,0.002671457],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.003949816,"threshold_uncertainty_score":0.01942277,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02828820071938948,"score_gpt":0.2923358372957938,"score_spread":0.2640476365764043,"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."}}