Research map / Architectures and efficiency / RNNs and normalization
Towards Understanding Regularization in 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 |
|---|---|---|---|
| Add dropout ratio 0.2 after each WN layer | WN · CIFAR10 validation | accuracy: 0.90 to 0.93 | Helped |
| Add dropout ratio 0.1 after each residual block | BN at batch size 1024 on ResNet18 · CIFAR10 validation | accuracy | Helped |
| Increase batch size from 128 to 1024 | BN on ResNet18 · CIFAR10 validation | accuracy | Hurt |
| Replace BN with WN | CNN · CIFAR10 | generalization error | Hurt |
| Use mean-only BN regularization (ablation) | WN · CIFAR10 | generalization error | Helped |
| Use variance-only BN regularization (ablation) | WN · CIFAR10 | generalization error | Helped |
| Replace BN with PN and add adaptive gamma decay | ResNet18 · CIFAR10 | validation accuracy | Helped |
| Remove batch mean and variance normalization (ablation) | BN-trained network, then fine-tuned with vanilla SGD · CIFAR10 | validation loss | Hurt |
| Increase batch size (ablation) | BN · CIFAR10 validation | validation loss | Hurt |
| Increase the maximum learning rate relative to WN or SGD | BN · Teacher-student optimization analysis | maximum learning rate: at least 2ζ larger | Helped |
Closest papers by meaning
- Understanding Batch Normalization (NeurIPS 2018 · similarity 0.94)
- Demystifying Batch Normalization in ReLU Networks: Equivalent Convex Optimization Models and Implicit Regularization (ICLR 2022 · similarity 0.92)
- Four Things Everyone Should Know to Improve Batch Normalization (ICLR 2020 · similarity 0.91)
- Normalizing the Normalizers: Comparing and Extending Network Normalization Schemes (ICLR 2017 · similarity 0.91)
- Towards Stabilizing Batch Statistics in Backward Propagation of Batch Normalization (ICLR 2020 · similarity 0.91)
- Ghost Noise for Regularizing Deep Neural Networks (AAAI 2024 · similarity 0.89)
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.