AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Softmax-Free Linear Transformers

Jiachen Lu, Junge Zhang, Xiatian Zhu, Jianfeng Feng, Tao Xiang, Li Zhang

IJCV 2024 · Architectures and efficiency › Attention and transformers · approach: Attention head and block redesign · 20 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 symmetric attention normalization (SOFT -> SOFT++) and test at larger inputSOFT-Small · ImageNet-1K val, 512x512 test inputtop-1 accuracy (%): 0.14 -> 80.2Improved
add attention normalization (SOFT -> SOFT++) at 384x384 testSOFT-Small · ImageNet-1K val, 384x384 test inputtop-1 accuracy (%): 33.7 -> 82.6Improved
add attention normalization to the Large modelSOFT-Large · ImageNet-1K valtop-1 accuracy (%): 83.6 -> 84.1Improved
add attention normalization to the Medium modelSOFT-Medium · ImageNet-1K valtop-1 accuracy (%): 83.2 -> 83.7Improved
add attention normalization to the Small modelSOFT-Small · ImageNet-1K valtop-1 accuracy (%): 82.5 -> 82.6No clear effect
use convolution sampling instead of biased sampling of first m tokens (ablation)SOFT++-Tiny · ImageNet-1K val (Tiny)top-1 accuracy (%): 79.9 -> 80.9Improved
use convolution sampling instead of random sampling (ablation)SOFT++-Tiny · ImageNet-1K val (Tiny)top-1 accuracy (%): 80.8 -> 80.9No clear effect
use convolution sampling instead of average pooling (ablation)SOFT++-Tiny · ImageNet-1K val (Tiny)top-1 accuracy (%): 80.8 -> 80.9No clear effect
use pyramidal multi-stage architecture (ablation)SOFT++-Small · ImageNet-1K val (Tiny)top-1 accuracy (%): 80.1 -> 82.4Improved
use overlapped convolution in patch embedding (ablation)SOFT++-Tiny · ImageNet-1K val (Tiny)top-1 accuracy (%): 78.4 -> 80.9Improved
increase bottleneck tokens from 36 to 49 (ablation)SOFT++-Tiny · ImageNet-1K val (Tiny)top-1 accuracy (%): 80.5 -> 80.9Improved
increase bottleneck tokens from 49 to 81 (ablation)SOFT++-Tiny · ImageNet-1K val (Tiny)top-1 accuracy (%): 80.9 -> 80.4Hurt

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