AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Always Skip Attention

Yiping Ji, Hemanth Saratchandran, Peyman Moghadam, Simon Lucey

ICCV 2025 · Architectures and efficiency › Attention and transformers · approach: Attention head and block redesign · 8 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
add skip connection to the self-attention block (vs removing it) (ablation)ViT-Tiny · CIFAR-10accuracy: approximately 22%Improved
add skip connection to the FFN (vs removing it) (ablation)ViT-Tiny · CIFAR-10accuracy: 2%Improved
add skip connections to ConvMixer-Tiny (ablation)ConvMixer-Tiny · three datasetsaccuracy: less than ±0.2%No clear effect
add DCT token grayingViT-Tiny with softmax attention · Tiny-ImageNettop-1 accuracy: 43.0 -> 44.8Improved
add SVD token grayingViT-Tiny with softmax attention · Tiny-ImageNettop-1 accuracy: 43.0 -> 44.7Improved
add DCT token grayingViT-Tiny with linear attention · Tiny-ImageNettop-5 accuracy: 62.7 -> 68.8Improved
add DCT token grayingViT-Base · ImageNet-1Ktop-1 accuracy: 81.0 -> 81.3Improved
add DCT token grayingSwin-S · ImageNet-1Ktop-1 accuracy: 81.3 -> 81.6Improved
add DCT token grayingCaiT-S · ImageNet-1Ktop-1 accuracy: 82.6 -> 82.7Improved
add DCT token graying to MAE pretrainingMAE ViT-Base · ImageNet-1K finetunetop-1 accuracy: 83.0 -> 83.2Improved
lower epsilon for SVD token graying from 0.7 to 0.5 (ablation)ViT-Base + SVDTG · ImageNet-1Ktop-1 accuracy: 81.4 -> 81.0Hurt
use DCT instead of SVD for token grayingViT-Base · ImageNet-1Ktraining time (days): 4.552 -> 0.732Improved

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