{"paper":{"title":"Maximizing Capacity Utilization of LoRA Adapters for Mitigating the Learning-Forgetting Trade-off","arxiv_id":"2611.17191","primary_cat":"cs.LG","submitted_at":"2026-11-12T17:55:33+00:00","version_compiled":"v1"},"notice":"Every quoted sentence is the authors' text from the arXiv listing abstract. Labels and probabilities are the outputs of classifier-1.13.0 on the question bank abstract-p1-20260923, exactly as returned. No sentence in this record was written by a language model.","schema":"pith.compiled-record/v0","status":"compiled","abstract":{"source":"arXiv listing abstract v1","sentences":[{"i":0,"p":1.0,"id":"s0","own":false,"role":"background","text":"Low-Rank Adaptation (LoRA) often underperforms full fine-tuning, especially in heterogeneous and continual-learning settings in which substantial forgetting occurs.","hedge":{"p":0.99,"label":"asserted"},"own_p":0.08,"main_claim_p":0.03,"quantitative":true,"quantitative_p":0.73,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.0,"background":1.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":1,"p":1.0,"id":"s1","own":false,"role":"background","text":"In general, it suffers from an undesirable learning-forgetting trade-off across diverse fine-tuning settings.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.1,"main_claim_p":0.04,"quantitative":false,"quantitative_p":0.06,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.0,"background":1.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":2,"p":0.51,"id":"s2","own":true,"role":"result","text":"We identify a potential contributor: LoRA adapters, especially high-rank ones, frequently fail to realize the available capacity afforded by the configured rank.","hedge":{"p":0.9,"label":"hedged_possibility"},"own_p":0.72,"main_claim_p":0.11,"quantitative":false,"quantitative_p":0.11,"role_probabilities":{"gap":0.27,"method":0.0,"result":0.51,"background":0.01,"limitation":0.0,"future_work":0.0,"contribution":0.19,"interpretation":0.02}},{"i":3,"p":0.97,"id":"s3","own":true,"role":"result","text":"Specifically, we find that standard LoRA updates have effective rank far below their configured rank, limiting how broadly the update energy is distributed.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.94,"main_claim_p":0.07,"quantitative":true,"quantitative_p":0.57,"role_probabilities":{"gap":0.01,"method":0.0,"result":0.97,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.02,"interpretation":0.0}},{"i":4,"p":0.9,"id":"s4","own":true,"role":"contribution","text":"Therefore, we propose training mechanisms that optimize the original objective while maximizing adapter effective rank.","hedge":{"p":0.99,"label":"asserted"},"own_p":0.59,"main_claim_p":0.19,"quantitative":false,"quantitative_p":0.04,"role_probabilities":{"gap":0.0,"method":0.1,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.9,"interpretation":0.0}},{"i":5,"p":0.95,"id":"s5","own":true,"role":"method","text":"Our main method adds a simple, stable, differentiable, and SVD-free R\\'enyi-2 effective-rank regularizer to the language-modeling loss.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.94,"main_claim_p":0.1,"quantitative":false,"quantitative_p":0.08,"role_probabilities":{"gap":0.0,"method":0.95,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.05,"interpretation":0.0}},{"i":6,"p":0.57,"id":"s6","own":true,"role":"contribution","text":"We also connect capacity maximization to norm-bounded updates.","hedge":{"p":0.98,"label":"asserted"},"own_p":0.78,"main_claim_p":0.04,"quantitative":false,"quantitative_p":0.11,"role_probabilities":{"gap":0.0,"method":0.39,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.57,"interpretation":0.04}},{"i":7,"p":1.0,"id":"s7","own":true,"role":"result","text":"In mixed-domain rank sweeps on three models evaluated with multiple random seeds, a penalty weight of $\\lambda=0.01$ lowers mean Wiki perplexity at every tested rank of at least $16$ relative to unpenalized LoRA.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.97,"main_claim_p":0.12,"quantitative":true,"quantitative_p":0.99,"role_probabilities":{"gap":0.0,"method":0.0,"result":1.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":8,"p":0.99,"id":"s8","own":true,"role":"result","text":"Mean training perplexity improves or increases by at most $0.02$ in all but one model-rank comparison.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.95,"main_claim_p":0.04,"quantitative":true,"quantitative_p":0.99,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.99,"background":0.0,"limitation":0.01,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":9,"p":0.98,"id":"s9","own":true,"role":"result","text":"Single-task and scheduling experiments further show that the balance between learning and retention depends on the configuration.","hedge":{"p":0.98,"label":"asserted"},"own_p":0.91,"main_claim_p":0.06,"quantitative":false,"quantitative_p":0.06,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.98,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.02}},{"i":10,"p":0.41,"id":"s10","own":true,"role":"interpretation","text":"We also observe that several LoRA variants in the literature also suffer from the identified capacity issue, and our approach provides a complementary axis of improvement to LoRA-based parameter-efficient fine-tuning.","hedge":{"p":0.97,"label":"asserted"},"own_p":0.69,"main_claim_p":0.09,"quantitative":false,"quantitative_p":0.22,"role_probabilities":{"gap":0.01,"method":0.0,"result":0.32,"background":0.03,"limitation":0.0,"future_work":0.0,"contribution":0.23,"interpretation":0.41}},{"i":11,"p":1.0,"id":"s11","own":true,"role":"interpretation","text":"These results suggest that effective rank is not only a post-hoc diagnostic of adapter quality, but a trainable objective for improving the learning-forgetting frontier.","hedge":{"p":0.99,"label":"hedged_possibility"},"own_p":0.69,"main_claim_p":0.81,"quantitative":false,"quantitative_p":0.06,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":1.0}},{"i":12,"p":0.69,"id":"s12","own":false,"role":"contribution","text":"Our code is available at https://github.com/tomohiro-sawada/lora-capacity-utilization","hedge":{"p":0.89,"label":"asserted"},"own_p":0.44,"main_claim_p":0.02,"quantitative":false,"quantitative_p":0.04,"role_probabilities":{"gap":0.0,"method":0.05,"result":0.0,"background":0.02,"limitation":0.0,"future_work":0.24,"contribution":0.69,"interpretation":0.0}}]},"compiled":{"main_claim":{"p":1.0,"text":"These results suggest that effective rank is not only a post-hoc diagnostic of adapter quality, but a trainable objective for improving the learning-forgetting frontier.","location":"abstract s11","hedge_strength":{"p":0.99,"label":"hedged possibility"}},"provenance":{"route":{"via":"TypeSafe API","model":"classifier-1.13.0","endpoint":"https://api.typesafe.ai/v1/systemone","provider":"TypeSafe"},"packing":"one request per paper","requests":1,"input_tokens":9562,"cost_usd_reported":0.0},"compiled_at":"2026-11-14","bank_version":"abstract-p1-20260923","model_version":"classifier-1.13.0","abstract_states":{"claims_first":0.05,"states_limitation":0.44,"human_participants":0.02,"controlled_comparison":0.93,"randomized_assignment":0.22,"states_quantitative_result":0.99,"compares_with_prior_results":0.95,"mentions_code_or_data_availability":0.99},"evidential_basis":{"p":0.95,"value":"observation or experiment","probabilities":{"argument":0.0,"literature":0.0,"proof_or_derivation":0.0,"analysis_of_existing_data":0.0,"observation_or_experiment":0.95,"simulation_or_computation":0.05}},"contribution_type":{"p":1.0,"value":"method or tool","probabilities":{"review":0.0,"method_or_tool":1.0,"empirical_study":0.0,"simulation_study":0.0,"theoretical_result":0.0,"dataset_or_benchmark":0.0,"position_or_commentary":0.0}},"contribution_statement":{"p":0.81,"text":"Our main method adds a simple, stable, differentiable, and SVD-free R\\'enyi-2 effective-rank regularizer to the language-modeling loss.","location":"abstract s5"}},"coverage":{"tier":"abstract","levels_run":["P abstract"],"not_examined":["full text","figures and tables","references and citing sentences","relations to other papers"],"sources_read":["arXiv listing abstract v1"],"levels_not_run":["S sections","T body sentences","C citations","R cross-paper relations"],"sentences_examined":13,"sentences_in_abstract":13},"measured_accuracy":{"url":"https://pith.science/accuracy","measured_on":"2026-10-02","method":"two independent model families labelled a fresh random sample blind; the figure is how often the published label matches their consensus","sentence_role_agreement":0.92,"main_claim_agreement":0.908,"typed_citation_agreement":0.933,"typed_citation_agreement_95_interval":[0.917,0.945],"review_comments_accurate":0.758,"review_comments_accurate_95_interval":[0.684,0.82],"review_caught_admitted_error_found_or_related":0.448,"review_comments_judged":{"judged_by":["gpt-5.6-luna","grok-4.7"],"judged_against":"Pith's own text copy of each paper: text extracted from the PDF, the first 90,000 characters","measured_on":"2026-10-04","accurate":113,"decided":149,"sampled_reviews":200,"split_by_lane":false,"by_lane":"https://pith.science/accuracy#by-lane"}}}