English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Batch normalization (BN) is a popular and ubiquitous method in deep learningthat has been shown to decrease training time and improve generalizationperformance of neural networks. Despite its success, BN is not theoreticallywell understood. It is not suitable for use with very small mini-batch sizes oronline learning. In this paper, we propose a new method called BatchNormalization Preconditioning (BNP). Instead of applying normalizationexplicitly through a batch normalization layer as is done in BN, BNP appliesnormalization by conditioning the parameter gradients directly during training.This is designed to improve the Hessian matrix of the loss function and henceconvergence during training. One benefit is that BNP is not constrained on themini-batch size and works in the online learning setting. Furthermore, itsconnection to BN provides theoretical insights on how BN improves training andhow BN is applied to special architectures such as convolutional neuralnetworks. For a theoretical foundation, we also present a novel Hessiancondition number based convergence theory for a locally convex but notstrong-convex loss, which is applicable to networks with a scale-invariantproperty.

Batch Normalization Preconditioning for Neural Network Training