AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

Parameter-Efficient Model Adaptation for Vision Transformers

Xuehai He, Chunyuan Li, Pengchuan Zhang, Jianwei Yang, Xin Eric Wang

AAAI 2023 · Architectures and efficiency › Pruning and efficient tuning · approach: Efficient ViT adaptation · 71 citations (OpenAlex)

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
use KAdaptation rather than the strongest efficient baseline, LoRACLIP ViT-B-224/32 · ELEVATER 20 datasets, 5-shotaverage accuracy (%): 66.32 -> 68.92Improved
use KAdaptation rather than LoRACLIP · seven full-shot image-classification datasetsaverage accuracy (%): 79.3 -> 79.6Improved
use KAdaptation rather than LoRASupervised ViT · seven full-shot image-classification datasetsaverage accuracy (%): 78.6 -> 79.2Improved
use KAdaptation instead of full-model fine-tuningCLIP · seven full-shot image-classification datasetsaverage accuracy (%): 80.6 -> 79.6Hurt
apply KAdaptation to MLP modules instead of attention modules (ablation)KAdaptation · CIFAR10, CIFAR100, SUN397average accuracy (%): 88.1 -> 86.6Hurt
place Adapters in attention layers rather than using standard Adapter placement (ablation)Adapter-tuning · CIFAR10, CIFAR100, SUN397average accuracy (%): 87.7 -> 54.1Hurt
select attention modules based on their lower local intrinsic dimensionSupervised ViT · CIFAR100, first-layer component intrinsic-dimension experimentlocal intrinsic dimension: 575 -> 300Improved

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