AI Paper Atlas

Research map / Optimization / Implicit bias and generalization

Batch Normalization Preconditioning for Neural Network Training

Susanna Lange, Kyle Helfrich, Qiang Ye

JMLR 2022 · Optimization › Implicit bias and generalization · approach: Stability-based generalization analysis

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
use BNP instead of BN at mini-batch size 6fully connected network · CIFAR10 FC (batch 6)test accuracyImproved
use BNP instead of BN at mini-batch size 25-layer CNN · CIFAR10 CNN (batch 2)test accuracyImproved
use BNP at mini-batch size 1 where BN does not train5-layer CNN · CIFAR10 CNN (batch 1)test accuracyImproved
use BNP instead of BN with large mini-batch (60)fully connected network · MNIST FC (batch 60)test accuracyNo clear effect
use BNP instead of BN with large mini-batch (60)fully connected network · CIFAR10 FC (batch 60)test accuracyMixed
use BNP instead of BN at mini-batch size 1285-layer CNN · CIFAR10 CNN (batch 128)test accuracyNo clear effect
add BNP preconditioning to a vanilla networkfully connected network · CIFAR10 FC (batch 60)Hessian condition numberImproved
add BNP to a vanilla networkfully connected network · CIFAR10 FCtest accuracy and training lossImproved
add GN to BNP on ResNet-110 (BNP+GN vs BNP alone) (ablation)ResNet-110 · CIFAR10test accuracy at learning rate decayImproved
use BNP+GN instead of BN, before first learning rate decayResNet-110 · CIFAR10test accuracyImproved
use BNP+GN instead of BN, after learning rate decayPreactivation ResNet-110 · CIFAR100test accuracyHurt
use BNP+GN instead of BN, after second decayResNet-18 · ImageNettest accuracyHurt

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