{"id":"W4407184900","doi":"10.48550/arxiv.2502.01755","title":"Robust Federated Finetuning of LLMs via Alternating Optimization of LoRA","year":2025,"lang":"en","type":"preprint","venue":"ArXiv.org","topic":"Advanced Numerical Analysis Techniques","field":"Engineering","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Government of Canada; Canadian Institute for Advanced Research","keywords":"Computer science; Business","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001936062,0.001437738,0.001403295,0.0006198623,0.0005746986,0.001396699,0.001655115,0.001585949,0.005173657],"category_scores_gemma":[0.01086832,0.0006925199,0.0007647222,0.0004073234,0.001237083,0.001991129,0.002354997,0.002475987,0.001695558],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000707922,"about_ca_system_score_gemma":0.001603867,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003978414,"about_ca_topic_score_gemma":0.007475148,"domain_scores_codex":[0.9990079,0.0003914953,0.00005337353,0.0002213742,0.0002016416,0.0001242733],"domain_scores_gemma":[0.9972245,0.001649858,0.0001881135,0.0004504246,0.0003457015,0.000141543],"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":[0.0002407213,0.0001426169,0.0008299979,0.000125426,0.00008159911,0.00009207259,0.0001129724,0.8469169,0.005640931,0.007963357,0.003666649,0.1341866],"study_design_scores_gemma":[0.00001279222,0.00002512106,0.00004256951,0.000006399768,0.000004212858,0.00001178917,0.000008217943,0.9956994,0.000690076,0.003166543,0.0003273973,0.000005420732],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.02439077,0.0003403242,0.9674284,0.0003005644,0.00007314995,0.00006640482,0.00008079127,0.00388646,0.003432987],"genre_scores_gemma":[0.7044143,0.0001437465,0.2878454,0.0006428099,0.00009171954,0.0003523566,0.0002707067,0.0007812983,0.005457648],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.005173657,"threshold_uncertainty_score":0.01730758,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0328022328642046,"score_gpt":0.2596119639063957,"score_spread":0.2268097310421911,"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."}}