AI Paper Atlas

Research map / Detection and segmentation / CNN features

Is normalization indispensable for training deep neural network?

Jie Shao, Kai Hu, Changhu Wang, Xiangyang Xue, Bhiksha Raj

NeurIPS 2020 · Detection and segmentation › CNN features · approach: Efficient network training and design · 28 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
replace no multiplier with a learnable scalar multiplier per residual block (ablation)RescaleNet · ImageNet valtop-1 accuracy (%): 74.2 -> 76.6Improved
use scalar rather than vector multipliers (ablation)RescaleNet · ImageNet valtop-1 accuracy (%): 74.8 -> 76.6Improved
set hyper-parameter c = L instead of c = 1 or L2 (ablation)RescaleNet · ImageNet valtop-1 accuracy (%): 76.4 -> 76.6No clear effect
data-dependent bias initialization instead of zero init (ablation)RescaleNet · ImageNet valtop-1 accuracy (%): 76.4 -> 76.6Improved
apply bias before weight multiplication instead of after (ablation)RescaleNet · ImageNet valtop-1 accuracy (%): 76.1 -> 76.6Improved
add dropout to non-normalized RescaleNetResNet50 · ImageNet valtop-1 accuracy (%): 74.3 -> 76.6Improved
RescaleNet instead of Group NormalizationResNet50 · ImageNet valtop-1 accuracy (%): 75.9 -> 76.58Improved
RescaleNet instead of Batch Normalization with 200-epoch refined recipeResNet-D · ImageNet valtop-1 accuracy (%): 79.15 -> 78.93No clear effect
RescaleNet instead of BN on SE-ResNet50SE-ResNet50 · ImageNet valtop-1 accuracy (%): 76.7 -> 77.22Improved
RescaleNet instead of Group Norm on Mask R-CNN FPN 3XResNet50 · COCO val2017AP bbox: 40.8 -> 41.0Improved
RescaleNet instead of Layer Norm in transformerTransformer · IWSLT14 DE-ENBLEU: 34.2 -> 35.0Improved
pre-weight bias only, without normalization, on VGG-19VGG-19 · ImageNet valtop-1 accuracy (%): 74.4 -> 75.0Improved

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