AI Paper Atlas

Research map / Detection and segmentation / CNN features

Rectifying Magnitude Neglect in Linear Attention

Qihang Fan, Huaibo Huang, Yuang Ai, Ran He

ICCV 2025 · Detection and segmentation › CNN features · approach: Convolution-transformer backbones · 22 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
discard Q magnitude by Q/||Q|| in softmax attention (ablation)DeiT-T softmax · ImageNet-1Ktop-1 accuracy: 72.2 -> 70.0Hurt
add magnitude-aware beta (ablation)MAViT-T · ImageNet-1Ktop-1 accuracy: 52.3 -> 82.9Improved
add gamma offset (ablation)MAViT-T · ImageNet-1Ktop-1 accuracy: NaN -> 82.9Improved
fixed beta,gamma instead of learnable (ablation)MAViT-T · ImageNet-1Ktop-1 accuracy: 71.7 -> 82.9Improved
swap kernel ELU+1 for ReLU (ablation)MAViT-T · ImageNet-1Ktop-1 accuracy: 82.9 -> 82.8No clear effect
swap kernel ELU+1 for exp (ablation)MAViT-T · ADE20KmIoU: 47.6 -> 47.4No clear effect
replace InLine attention with MALASwin-S · ImageNet-1Ktop-1 accuracy: 83.6 -> 85.3Improved
replace InLine attention with MALADeiT-T · ImageNet-1Ktop-1 accuracy: 74.5 -> 75.1Improved
replace vanilla linear attention with MALAConformer(S) · LibriSpeech test-otherWER: 10.2 -> 5.3Improved

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