Research map / Architectures and efficiency / RNNs and normalization
How Does Batch Normalization Help Optimization?
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 BatchNorm | VGG-like network · CIFAR-10 test | accuracy: 80% -> 83% | Helped |
| Add BatchNorm | VGG-like network · CIFAR-10 test with data augmentation | accuracy: 88% -> 92% | Helped |
| Add time-varying, non-zero-mean, non-unit-variance noise after BatchNorm (ablation) | BatchNorm VGG · CIFAR-10 | training performance | No clear effect |
| Add time-varying activation noise after BatchNorm (ablation) | VGG without BatchNorm · CIFAR-10 | training performance | Hurt |
| Replace BatchNorm with L1 normalization (ablation) | VGG-like network · CIFAR-10 | training performance | Helped |
| Replace BatchNorm with L2 normalization (ablation) | VGG-like network · CIFAR-10 | training performance | Helped |
| Replace BatchNorm with L-infinity normalization (ablation) | VGG-like network · CIFAR-10 | training performance | Helped |
| Replace BatchNorm with L1 normalization (ablation) | 25-layer deep linear network · Synthetic Gaussian data | training performance | Helped |
| Replace BatchNorm with L2 or L-infinity normalization (ablation) | 25-layer deep linear network · Synthetic Gaussian data | training performance | Helped |
| Add BatchNorm | VGG and deep linear networks · CIFAR-10 and synthetic Gaussian data | internal covariate shift | Hurt |
Closest papers by meaning
- Beyond BatchNorm: Towards a Unified Understanding of Normalization in Deep Learning (NeurIPS 2021 · similarity 0.92)
- Deconstructing the Regularization of BatchNorm (ICLR 2021 · similarity 0.92)
- Understanding Batch Normalization (NeurIPS 2018 · similarity 0.92)
- Four Things Everyone Should Know to Improve Batch Normalization (ICLR 2020 · similarity 0.91)
- Towards Training Without Depth Limits: Batch Normalization Without Gradient Explosion (ICLR 2024 · similarity 0.89)
- Reducing Divergence in Batch Normalization for Domain Adaptation (AAAI 2025 · similarity 0.88)
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.