AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

MixA: A Mixed Attention approach with Stable Lightweight Linear Attention to enhance Efficiency of Vision Transformers at the Edge

Sabbir Ahmed, Jingtao Li, Weiming Zhuang, Chen Chen, Lingjuan Lyu

ICCV 2025 · Architectures and efficiency › Attention and transformers · approach: Hybrid and interpretable ViT designs · 2 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
replace softmax attention with MixA (ReLU quadratic in critical layers, SteLLA linear elsewhere)DeiT-T · ImageNet-1K valtop-1 accuracy (%): 74.66 -> 74.54No clear effect
replace softmax attention with MixASwin-S · ImageNet-1K valtop-1 accuracy (%): 83.59 -> 82.88Hurt
replace softmax attention with MixASwin-T · COCO valmAP (%): 34.6 -> 32.6Hurt
replace softmax attention with MixASwin-T · ADE20KmIoU (%): 35.3 -> 35.8Improved
use SteLLA normalization instead of CosFormer row-sum normalization in ReLU linear attention (ablation)DeiT-T all-linear attention · ImageNet-1K valtop-1 accuracy (%): 69.05 -> 73.71Improved
add targeted quadratic attention (k=6) to SteLLA (ablation)SteLLA DeiT-T · ImageNet-1K valtop-1 accuracy (%): 73.71 -> 74.54Improved
choose quadratic-attention layers by importance score instead of randomly (ablation)MixA DeiT-T · ImageNet-1K valtop-1 accuracy (%): 73.98 -> 74.54Improved
add learnable scaling parameters (ablation)MixA DeiT-T · ImageNet-1K valtop-1 accuracy (%): 74.44 -> 74.54Improved
increase number of quadratic layers k from 6 to 9 (ablation)MixA DeiT-T · ImageNet-1K valtop-1 accuracy (%): 74.54 -> 74.54No clear effect
increase number of quadratic layers k from 3 to 6 (ablation)MixA DeiT-T · ImageNet-1K valtop-1 accuracy (%): 74.28 -> 74.54Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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