Not All Directions Matter: Towards Structured and Task-Aware Low-Rank Model Adaptation

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Xiao, Xi, Ma, Chenrui, Zhang, Yunbei, Liu, Chen, Wang, Zhuxuanzi, Li, Yanshu, Zhao, Lin, Hu, Guosheng, Wang, Tianyang, Xu, Hao
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918467322511360
author Xiao, Xi
Ma, Chenrui
Zhang, Yunbei
Liu, Chen
Wang, Zhuxuanzi
Li, Yanshu
Zhao, Lin
Hu, Guosheng
Wang, Tianyang
Xu, Hao
author_facet Xiao, Xi
Ma, Chenrui
Zhang, Yunbei
Liu, Chen
Wang, Zhuxuanzi
Li, Yanshu
Zhao, Lin
Hu, Guosheng
Wang, Tianyang
Xu, Hao
contents Low-Rank Adaptation (LoRA) has become a cornerstone of parameter-efficient fine-tuning (PEFT). Yet, its efficacy is hampered by two fundamental limitations: semantic drift, by treating all update directions with equal importance, and structural incoherence, from adapting layers independently, resulting in suboptimal, uncoordinated updates. To remedy these, we propose StructLoRA, a framework that addresses both limitations through a principled, dual-component design: (1) an Information Bottleneck-guided filter that prunes task-irrelevant directions to mitigate semantic drift, and (2) a lightweight, training-only graph-based coordinator that enforces inter-layer consistency to resolve structural incoherence. Extensive experiments across large language model , vision language model, and vision model (including LLaMA, LLaVA, and ViT) demonstrate that StructLoRA consistently establishes a new state-of-the-art, outperforming not only vanilla LoRA but also advanced dynamic rank allocation and sparsity-based methods. Notably, the benefits are particularly pronounced in challenging low-rank and low-data regimes. Crucially, since our proposed modules operate only during training, StructLoRA enhances performance with zero additional inference cost, advancing the focus of PEFT -- from mere parameter compression to a more holistic optimization of information quality and structural integrity.
format Preprint
id arxiv_https___arxiv_org_abs_2603_14228
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Not All Directions Matter: Towards Structured and Task-Aware Low-Rank Model Adaptation
Xiao, Xi
Ma, Chenrui
Zhang, Yunbei
Liu, Chen
Wang, Zhuxuanzi
Li, Yanshu
Zhao, Lin
Hu, Guosheng
Wang, Tianyang
Xu, Hao
Computer Vision and Pattern Recognition
Low-Rank Adaptation (LoRA) has become a cornerstone of parameter-efficient fine-tuning (PEFT). Yet, its efficacy is hampered by two fundamental limitations: semantic drift, by treating all update directions with equal importance, and structural incoherence, from adapting layers independently, resulting in suboptimal, uncoordinated updates. To remedy these, we propose StructLoRA, a framework that addresses both limitations through a principled, dual-component design: (1) an Information Bottleneck-guided filter that prunes task-irrelevant directions to mitigate semantic drift, and (2) a lightweight, training-only graph-based coordinator that enforces inter-layer consistency to resolve structural incoherence. Extensive experiments across large language model , vision language model, and vision model (including LLaMA, LLaVA, and ViT) demonstrate that StructLoRA consistently establishes a new state-of-the-art, outperforming not only vanilla LoRA but also advanced dynamic rank allocation and sparsity-based methods. Notably, the benefits are particularly pronounced in challenging low-rank and low-data regimes. Crucially, since our proposed modules operate only during training, StructLoRA enhances performance with zero additional inference cost, advancing the focus of PEFT -- from mere parameter compression to a more holistic optimization of information quality and structural integrity.
title Not All Directions Matter: Towards Structured and Task-Aware Low-Rank Model Adaptation
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2603.14228