Research map / Architectures and efficiency / RNNs and normalization
Proxy-Normalizing Activations to Match Batch Normalization while Removing Batch Dependence
Antoine Labatie, Dominic Masters, Zach Eaton-Rosen, Carlo Luschi
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 Proxy Normalization to LN (with extra regularization) (ablation) | Layer Normalization · ResNet-50 ImageNet | ImageNet val accuracy (%): 74.6 -> 76.5 | Improved |
| add Proxy Normalization to LN (without extra regularization) (ablation) | Layer Normalization · ResNet-50 ImageNet | ImageNet val accuracy (%): 74.5 -> 75.9 | Improved |
| add PN to GN with 8 groups (ablation) | Group Normalization · ResNet-50 ImageNet | ImageNet val accuracy (%): 75.4 -> 76.7 | Improved |
| add PN to GN with 32 groups (ablation) | Group Normalization · ResNet-50 ImageNet | ImageNet val accuracy (%): 75.3 -> 76.1 | Improved |
| add PN to GN+WS (ablation) | GN+WS · ResNet-50 ImageNet | ImageNet val accuracy (%): 76.7 -> 77.1 | Improved |
| add PN to Batch Normalization (ablation) | BN · ResNet-50 ImageNet | ImageNet val accuracy (%): 76.3 -> 76.2 | No clear effect |
| replace BN with GN+PN | ResNet-101 · ImageNet | ImageNet val accuracy (%): 78.0 -> 78.6 | Improved |
| replace BN with GN+PN | ResNet-50 · ImageNet | ImageNet val accuracy (%): 75.8 -> 76.7 | Improved |
| replace BN with GN+PN | ResNeXt-101 · ImageNet | ImageNet val accuracy (%): 78.9 -> 79.0 | No clear effect |
| replace BN with GN+PN | EfficientNet-B2 (depthwise) · ImageNet | ImageNet val accuracy (%): 80.0 -> 80.0 | No clear effect |
| add PN to GN (ablation) | GN · EfficientNet-B0 (depthwise) ImageNet | ImageNet val accuracy (%): 76.2 -> 77.0 | Improved |
| add weight standardization to GN (ablation) | GN · EfficientNet-B0 (depthwise) ImageNet | ImageNet val accuracy (%): 76.2 -> 74.1 | Hurt |
Closest papers by meaning
- Normalizing the Normalizers: Comparing and Extending Network Normalization Schemes (ICLR 2017 · similarity 0.89)
- Beyond BatchNorm: Towards a Unified Understanding of Normalization in Deep Learning (NeurIPS 2021 · similarity 0.88)
- Filter Response Normalization Layer: Eliminating Batch Dependence in the Training of Deep Neural Networks (CVPR 2020 · similarity 0.88)
- Towards Training Without Depth Limits: Batch Normalization Without Gradient Explosion (ICLR 2024 · similarity 0.88)
- Batch Normalization Is Blind to the First and Second Derivatives of the Loss (AAAI 2024 · similarity 0.87)
- Reducing Divergence in Batch Normalization for Domain Adaptation (AAAI 2025 · similarity 0.86)
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.
