AI Paper Atlas

Research map / Architectures and efficiency / Neural architecture search

ElasticViT: Conflict-aware Supernet Training for Deploying Fast Vision Transformer on Diverse Mobile Devices

Chen Tang, Li Lyna Zhang, Huiqiang Jiang, Jiahang Xu, Ting Cao, Quanlu Zhang, Yuqing Yang, Zhi Wang, Mao Yang

ICCV 2023 · Architectures and efficiency › Neural architecture search · approach: One-shot and few-shot supernets · 30 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 adjacent step (complexity-aware) sampling (ablation)ElasticViT supernet · ImageNet supernet, 100 MFLOPstop-1 accuracy: 69.9 -> 73.2Improved
add adjacent step sampling at large FLOPs (ablation)ElasticViT supernet · ImageNet supernet, 800 MFLOPstop-1 accuracy: 79.0 -> 79.0No clear effect
add multiple hierarchical smallest subnets (HSS) (ablation)ElasticViT supernet · ImageNet supernet, 300 MFLOPstop-1 accuracy: 76.7 -> 78.4Improved
add multiple hierarchical smallest subnets (HSS) (ablation)ElasticViT supernet · ImageNet supernet, 100 MFLOPstop-1 accuracy: 73.2 -> 72.8Hurt
add performance-aware sampling (ablation)ElasticViT supernet · ImageNet supernet, 100 MFLOPstop-1 accuracy: 72.8 -> 73.8Improved
add all ElasticViT sampling techniques versus sandwich rule (ablation)ElasticViT supernet · ImageNet supernet, 200 MFLOPstop-1 accuracy: 74.2 -> 77.2Improved
apply conflict-aware sampling to reduce gradient conflicts (ablation)ElasticViT supernet · Gradient cosine similarity, 200 MFLOPs random subnetscosine similarity: 0.32 -> 0.51Improved
inherit supernet weights instead of retraining from scratch (ablation)searched ViTs · ImageNet, 200 MFLOPstop-1 accuracy: 75.3 -> 77.2Improved
inherit supernet weights instead of retraining from scratch (ablation)searched ViTs · ImageNet, 700 MFLOPstop-1 accuracy: 79.9 (scratch) vs 79.8No clear effect
use FLOPs-focused hybrid ViTs such as MobileFormer instead of mobile-friendly opsmobile ViTs · Pixel4 latencylatency: MobileFormer 415.1 ms vs ElasticViT-S1 21.0 msHurt

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