Research map / Architectures and efficiency / Attention and transformers
VanillaNet: the Power of Minimalism in Deep Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 series activation function (n=1 vs plain ReLU) (ablation) | VanillaNet-6 · ImageNet-1k val | top-1 accuracy: 60.53 -> 74.53 | Helped |
| increase series count n from 1 to 3 (ablation) | VanillaNet-6 · ImageNet-1k val | top-1 accuracy: 74.53 -> 76.36 | Helped |
| increase series count n from 3 to 4 (ablation) | VanillaNet-6 · ImageNet-1k val | top-1 accuracy: 76.36 -> 76.43 | No clear effect |
| add deep training strategy (ablation) | VanillaNet-6 · ImageNet-1k val | top-1 accuracy: 75.23 -> 76.36 | Helped |
| add deep training and series activation (ablation) | AlexNet · ImageNet-1k val | top-1 accuracy: 57.52 -> 63.59 | Helped |
| add deep training and series activation (ablation) | ResNet-50 · ImageNet-1k val | top-1 accuracy: 76.13 -> 76.27 | No clear effect |
| add parameter-free shortcut before activation (ablation) | VanillaNet-6 · ImageNet-1k val | top-1 accuracy: 76.36 -> 75.92 | Hurt |
| add parameter-free shortcut after activation (ablation) | VanillaNet-6 · ImageNet-1k val | top-1 accuracy: 76.36 -> 75.72 | Hurt |
| replace Swin-T backbone with VanillaNet-13 | RetinaNet · COCO | AP^b: 41.5 -> 41.8 | Helped |
| replace Swin-T backbone with VanillaNet-13 | Mask RCNN · COCO | AP^b: 42.7 -> 42.9 | Helped |
| use series activation only (no deep training) vs plain baseline (ablation) | VanillaNet-6 · ImageNet-1k val | top-1 accuracy: 59.58 -> 75.23 | Helped |
Closest papers by meaning
- Paleo: A Performance Model for Deep Neural Networks (ICLR 2017 · similarity 0.87)
- Modular Networks: Learning to Decompose Neural Computation (NeurIPS 2018 · similarity 0.86)
- Foundation Models Secretly Understand Neural Network Weights: Enhancing Hypernetwork Architectures with Foundation Models (ICLR 2025 · similarity 0.86)
- Mechanistic Design and Scaling of Hybrid Architectures (ICML 2024 · similarity 0.85)
- LEMON: Lossless model expansion (ICLR 2024 · similarity 0.85)
- Random Sparse Lifts: Construction, Analysis and Convergence of finite sparse networks (ICLR 2024 · similarity 0.84)
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.