AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

LoPrune: Efficient Data Pruning for LoRA-Based Fine-Tuning of Vision Transformer

Qiang He, Yaozong Yang, Kaibin Wang, Ziteng Wei, Feifei Chen, Caslon Chua, Yun Yang

CVPR 2026 · Architectures and efficiency › Pruning and efficient tuning · approach: Efficient ViT adaptation

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
project influence scores onto the LoRA trainable subspace (TSAS) (ablation)LoPrune (K-FAC) · CIFAR-100accuracy: 86.88 -> 89.59Improved
project influence scores onto the LoRA trainable subspace (TSAS) (ablation)LoPrune (K-FAC) · Tiny-ImageNetaccuracy: 82.45 -> 84.78Improved
project influence scores onto the LoRA trainable subspace (TSAS) (ablation)LoPrune · CIFAR-100 (ViT-Small, pruning ratio 0.8)accuracy: 80.73 -> 88.65Improved
use K-FAC instead of pseudoinverse Hessian (ablation)LoPrune · CIFAR-100total time (s): 5542.3 -> 371.5Improved
use K-FAC instead of pseudoinverse Hessian (ablation)LoPrune · CIFAR-100accuracy: 89.64 -> 89.59No clear effect
use K-FAC instead of the Agarwal approximation (ablation)LoPrune · CIFAR-100accuracy: 89.13 -> 89.59Improved
start pruning later in fine-tuning (fully converged model for scoring) (ablation)LoPrune · ViT-Small, CIFAR-100 and Tiny-ImageNetaccuracy: max +0.22% accuracy, training time up to +68%No clear effect
LoPrune vs random pruning at 0.5 on COCO/classification (replace random with TSA scoring)DeiT-Tiny · CIFAR-100 (0.5 pruning)top-1 accuracy: 77.20 -> 79.88Improved
use warm-up-based scoring baselines (EL2N, GraNd) with few fine-tuning epochsEL2N / GraNd · CIFAR-100 and Tiny-ImageNet, high pruning ratioaccuracy: 5.08% - 10.34% lower than LoPrune above 0.8 pruningHurt
prune a small fraction (10%) of data with various pruning methodsViT-Small fine-tuning · CIFAR-100 and Tiny-ImageNetaccuracyImproved
fine-tune ResNets on a subset selected by LoPrune with ViT-Small instead of random subsetResNet-50 · CIFAR-100 (0.8 ratio retained)top-1 accuracy: 74.38 -> 77.83Improved
LoPrune at 0.5 pruning on Tiny-ImageNetViT-Small · Tiny-ImageNettotal training time (s): 992.1 -> 622.1Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy