Research map / Architectures and efficiency / Attention and transformers
Always Skip Attention
Yiping Ji, Hemanth Saratchandran, Peyman Moghadam, Simon Lucey
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 skip connection to the self-attention block (vs removing it) (ablation) | ViT-Tiny · CIFAR-10 | accuracy: approximately 22% | Improved |
| add skip connection to the FFN (vs removing it) (ablation) | ViT-Tiny · CIFAR-10 | accuracy: 2% | Improved |
| add skip connections to ConvMixer-Tiny (ablation) | ConvMixer-Tiny · three datasets | accuracy: less than ±0.2% | No clear effect |
| add DCT token graying | ViT-Tiny with softmax attention · Tiny-ImageNet | top-1 accuracy: 43.0 -> 44.8 | Improved |
| add SVD token graying | ViT-Tiny with softmax attention · Tiny-ImageNet | top-1 accuracy: 43.0 -> 44.7 | Improved |
| add DCT token graying | ViT-Tiny with linear attention · Tiny-ImageNet | top-5 accuracy: 62.7 -> 68.8 | Improved |
| add DCT token graying | ViT-Base · ImageNet-1K | top-1 accuracy: 81.0 -> 81.3 | Improved |
| add DCT token graying | Swin-S · ImageNet-1K | top-1 accuracy: 81.3 -> 81.6 | Improved |
| add DCT token graying | CaiT-S · ImageNet-1K | top-1 accuracy: 82.6 -> 82.7 | Improved |
| add DCT token graying to MAE pretraining | MAE ViT-Base · ImageNet-1K finetune | top-1 accuracy: 83.0 -> 83.2 | Improved |
| lower epsilon for SVD token graying from 0.7 to 0.5 (ablation) | ViT-Base + SVDTG · ImageNet-1K | top-1 accuracy: 81.4 -> 81.0 | Hurt |
| use DCT instead of SVD for token graying | ViT-Base · ImageNet-1K | training time (days): 4.552 -> 0.732 | Improved |
Closest papers by meaning
- Skip-Attention: Improving Vision Transformers by Paying Less Attention (ICLR 2024 · similarity 0.92)
- Token Selection is a Simple Booster for Vision Transformers (TPAMI 2022 · similarity 0.89)
- Attention is not all you need: pure attention loses rank doubly exponentially with depth (ICML 2021 · similarity 0.89)
- You Only Need Less Attention at Each Stage in Vision Transformers (CVPR 2024 · similarity 0.87)
- MLP Can Be A Good Transformer Learner (CVPR 2024 · similarity 0.87)
- On the Surprising Effectiveness of Attention Transfer for Vision Transformers (NeurIPS 2024 · 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.
