Research map / Architectures and efficiency / RNNs and normalization
Four Things Everyone Should Know to Improve Batch Normalization
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 |
|---|---|---|---|
| Use tuned example statistics in inference normalization | Batch Normalization on ImageNet with ResNet-152 · ImageNet ILSVRC 2012 validation | top-1 accuracy: 0.6% | Helped |
| Use Ghost Batch Normalization with tuned ghost batch size | Batch Normalization · Caltech-256 | accuracy: 5.8% | Helped |
| Use Ghost Batch Normalization with tuned ghost batch size | Batch Normalization · SVHN | accuracy | No clear effect |
| Add weight decay to normalization parameters γ and β | ResNet · CIFAR-100 | accuracy: 0.3% | Helped |
| Add weight decay to normalization parameters γ and β | Inception-v3 · Caltech-256 | accuracy: 0.13% | Hurt |
| Add weight decay to normalization parameters γ and β | ResNet-50 · Caltech-256 | accuracy: 0.91% | Helped |
| Add weight decay to normalization parameters γ and β | ResNet-18 · SVHN | accuracy: 98.79% | No clear effect |
| Combine Batch and Group Normalization and use inference example weighting | Group Normalization · CIFAR-100, B=2 | top-1 accuracy: 73.91% to 74.60%, 0.69% | Helped |
| Combine Batch and Group Normalization and use inference example weighting | Group Normalization · Caltech-256, B=2 | accuracy: 48.2% to 53.2%, 5.0% | Helped |
| Combine Batch and Group Normalization and use inference example weighting | Group Normalization · SVHN, B=2 | accuracy: 98.75% to 98.73% | Hurt |
| Apply the proposed normalization improvements | Batch Normalization · Flowers-102 | top-1 accuracy: 91.0% to 93.4%, 2.4% | Helped |
| Apply the proposed normalization improvements | Batch Normalization · CUB-2011 | top-1 accuracy: 81.1% to 82.4%, 1.4% | Helped |
Closest papers by meaning
- Understanding Batch Normalization (NeurIPS 2018 · similarity 0.93)
- Norm matters: efficient and accurate normalization schemes in deep networks (NeurIPS 2018 · similarity 0.92)
- Deconstructing the Regularization of BatchNorm (ICLR 2021 · similarity 0.92)
- MetaNorm: Learning to Normalize Few-Shot Batches Across Domains (ICLR 2021 · similarity 0.92)
- Ghost Noise for Regularizing Deep Neural Networks (AAAI 2024 · similarity 0.90)
- Reducing Divergence in Batch Normalization for Domain Adaptation (AAAI 2025 · similarity 0.90)
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.