AI Paper Atlas

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

NeurIPS 2021 · Architectures and efficiency › RNNs and normalization · approach: Normalization layer theory · 6 citations (OpenAlex)

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 changedWhereMeasuredEffect
add Proxy Normalization to LN (with extra regularization) (ablation)Layer Normalization · ResNet-50 ImageNetImageNet val accuracy (%): 74.6 -> 76.5Improved
add Proxy Normalization to LN (without extra regularization) (ablation)Layer Normalization · ResNet-50 ImageNetImageNet val accuracy (%): 74.5 -> 75.9Improved
add PN to GN with 8 groups (ablation)Group Normalization · ResNet-50 ImageNetImageNet val accuracy (%): 75.4 -> 76.7Improved
add PN to GN with 32 groups (ablation)Group Normalization · ResNet-50 ImageNetImageNet val accuracy (%): 75.3 -> 76.1Improved
add PN to GN+WS (ablation)GN+WS · ResNet-50 ImageNetImageNet val accuracy (%): 76.7 -> 77.1Improved
add PN to Batch Normalization (ablation)BN · ResNet-50 ImageNetImageNet val accuracy (%): 76.3 -> 76.2No clear effect
replace BN with GN+PNResNet-101 · ImageNetImageNet val accuracy (%): 78.0 -> 78.6Improved
replace BN with GN+PNResNet-50 · ImageNetImageNet val accuracy (%): 75.8 -> 76.7Improved
replace BN with GN+PNResNeXt-101 · ImageNetImageNet val accuracy (%): 78.9 -> 79.0No clear effect
replace BN with GN+PNEfficientNet-B2 (depthwise) · ImageNetImageNet val accuracy (%): 80.0 -> 80.0No clear effect
add PN to GN (ablation)GN · EfficientNet-B0 (depthwise) ImageNetImageNet val accuracy (%): 76.2 -> 77.0Improved
add weight standardization to GN (ablation)GN · EfficientNet-B0 (depthwise) ImageNetImageNet val accuracy (%): 76.2 -> 74.1Hurt

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy