{"id":"W4417003494","doi":"10.1109/icce-asia67487.2025.11263774","title":"Action Elimination through Hallucination Detection: A Reinforcement Learning Approach for LLM Fine-Tuning","year":2025,"lang":"","type":"article","venue":"","topic":"Adversarial Robustness in Machine Learning","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Nexen (Canada)","funders":"","keywords":"Reinforcement learning; Action (physics); Focus (optics); Filter (signal processing); Noise (video); Reinforcement","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":[0.002009167,0.001035437,0.0009907386,0.000461405,0.0003688288,0.0007436898,0.002118099,0.00107939,0.002154687],"category_scores_gemma":[0.007259306,0.000448765,0.0005011261,0.000215539,0.001284754,0.0010224,0.00178991,0.001924963,0.0005353183],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006092439,"about_ca_system_score_gemma":0.00105825,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00187578,"about_ca_topic_score_gemma":0.002127198,"domain_scores_codex":[0.9992658,0.0002474737,0.00004064937,0.0001997785,0.0001534757,0.00009283063],"domain_scores_gemma":[0.9975876,0.001366914,0.0003122193,0.0003175709,0.0002777047,0.0001380935],"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":[0.0002506757,0.0002374137,0.001394766,0.0001224067,0.00008638261,0.0001594293,0.0001752154,0.7725498,0.02232664,0.008993043,0.001316359,0.1923878],"study_design_scores_gemma":[0.00001304688,0.00004698622,0.0000721326,0.000005744365,0.000005191923,0.00001945771,0.000005385422,0.9950723,0.001697609,0.002800009,0.0002547965,0.000007305903],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.01419062,0.0001372643,0.9836274,0.0001440643,0.00003278177,0.00005755499,0.00001549754,0.0009476552,0.0008471286],"genre_scores_gemma":[0.8147039,0.00008023009,0.182238,0.0003545803,0.00006577769,0.0001876821,0.00005898951,0.0001747992,0.002136021],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.002154687,"threshold_uncertainty_score":0.0106256,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03738149689377482,"score_gpt":0.3173500053187909,"score_spread":0.2799685084250161,"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."}}