AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

SAL-ViT: Towards Latency Efficient Private Inference on ViT using Selective Attention Search with a Learnable Softmax Approximation

Yuke Zhang, Dake Chen, Souvik Kundu, Chenghao Li, Peter A. Beerel

ICCV 2023 · Architectures and efficiency › Attention and transformers · approach: Hybrid and interpretable ViT designs · 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
use external attention instead of self-attentionViT on CIFAR-10 · CIFAR-10approx PI latency (s): 81.48 -> 41.72Improved
use external attention instead of self-attentionViT on CIFAR-10 · CIFAR-10accuracy (%): 95.56 -> 92.86Hurt
replace softmax with scaling function in EAEA ViT · CIFAR-10accuracy (%): 92.86 -> 33.45Hurt
replace SA softmax with 2QuadSA ViT · CIFAR-10approx PI latency (s): 81.48 -> 51.12Improved
make L2Q parameters finer-grained (element-wise vs global) (ablation)L2Q · CIFAR-100accuracy (%): 76.08 -> 78.25Improved
make both shifting and scaling parameters learnable (ablation)element-wise L2Q · CIFAR-100accuracy (%): 77.6 -> 78.25Improved
select SA layers by learned alpha (SAS-L) rather than lowest alpha (worst case) (ablation)SAS · CIFAR-10/CIFAR-100/Tiny-ImageNetaccuracy (%)Improved
use head-wise search (SAS-H) instead of layer-wise (SAS-L) (ablation)SAS · CIFAR-100accuracy (%): 77.62 -> 77.15Hurt
use 2ReLU in SAS-L (ablation)SAS-L · Tiny-ImageNetaccuracy (%)Hurt
apply SAL-ViT hybrid attention with L2Q instead of full SA softmaxCCT ViT · ImageNetaccuracy (%): 77.22 -> 75.51 accuracy, 728.22 -> 450.87 s latencyMixed

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