Research map / Architectures and efficiency / Attention and transformers
EA-Vit: Efficient Adaptation for Elastic Vision Transformer
Chen Zhu, Wangbo Zhao, Huiwen Zhang, Yuhao Zhou, Weidong Tang, Shuo Wang, Zhihang Yuan, Yuzhang Shang, Xiaojiang Peng, Kai Wang, Dawei Yang
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add attention-head elasticity to MLP-only elasticity (ablation) | EA-ViT · CIFAR-100 | minimum MACs (G): 7.80 -> 6.04 | Improved |
| add attention-head elasticity to MLP-only elasticity (ablation) | EA-ViT · CIFAR-100 | AUC: 0.505 -> 0.646 | Improved |
| add embedding dimension elasticity (ablation) | EA-ViT · CIFAR-100 | minimum MACs (G): 6.04 -> 5.11 | Improved |
| add depth elasticity (ablation) | EA-ViT · CIFAR-100 | minimum MACs (G): 5.11 -> 4.60 | Improved |
| add depth elasticity (ablation) | EA-ViT · CIFAR-100 | max Top-1 accuracy: 93.31 -> 93.32 | No clear effect |
| add embedding elasticity (ablation) | EA-ViT · CIFAR-100 | AUC: 0.646 -> 0.651 | Improved |
| add all four elastic dimensions versus MLP only (ablation) | EA-ViT · CIFAR-100 | max Top-1 accuracy: 93.01 -> 93.32 | Improved |
| use router-selected submodels instead of manually selected fixed architectures (ablation) | EA-ViT · CIFAR-100 | MACs-accuracy curve | Improved |
| add Curriculum Elastic Adaptation (ablation) | EA-ViT · CIFAR-100 | MACs-accuracy curve | Mixed |
| add Curriculum Elastic Adaptation after Stage 2 refinement (ablation) | EA-ViT · CIFAR-100 | MACs-accuracy curve | Improved |
| add Pareto-optimal initialization of router instead of random (ablation) | EA-ViT · CIFAR-100 | MACs-accuracy curve | Improved |
| use EA-ViT instead of best prior method on Flowers-102 at 8 GMACs | EA-ViT · Flowers-102 | Top-1 accuracy: 73.80 -> 85.39 | Improved |
Closest papers by meaning
- Elastic ViTs from Pretrained Models without Retraining (NeurIPS 2025 · similarity 0.91)
- Slicing Vision Transformer for Flexible Inference (NeurIPS 2024 · similarity 0.89)
- Linearly Decomposing and Recomposing Vision Transformers for Diverse-Scale Models (NeurIPS 2024 · similarity 0.88)
- Dynamic Tuning Towards Parameter and Inference Efficiency for ViT Adaptation (NeurIPS 2024 · similarity 0.88)
- A General and Efficient Training for Transformer via Token Expansion (CVPR 2024 · similarity 0.87)
- Split Adaptation for Pre-trained Vision Transformers (CVPR 2025 · similarity 0.87)
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.
