Creative Commons license Spherical Perspective on Learning with Batch Normalization by Simon Roburrin (LIGM) [June 28, 2021]

 Summary

GdR ISIS Théorie du deep learning - June 28, 2021

Spherical Perspective on Learning with Batch Normalization

By Simon Roburrin (LIGM)

Batch Normalization (BN) is a prominent deep learning technique. In spite of its apparent simplicity, its implications over optimization are yet to be fully understood. In this paper, we introduce a spherical framework to study the optimization of neural networks with BN layers from a geometric perspective. More precisely, we leverage the radial invariance of groups of parameters, such as filters for convolutional neural networks, to translate the optimization steps on the L2 unit hypersphere. This formulation and the associated geometric interpretation shed new light on the training dynamics. Firstly, we use it to derive the first effective learning rate expression of Adam. Then we show that, in the presence of BN layers, performing SGD alone is actually equivalent to a variant of Adam constrained to the unit hypersphere. Finally, our analysis outlines phenomena that previous variants of Adam act on and we experimentally validate their importance in the optimization process.

This is a cowork by Simon Roburin, Yann de Mont-Marin, Andrei Bursuc, Renaud Marlet, Patrick Pérez, Mathieu Aubry

 Infos

  • Added by:

  • Contributor(s):

    • Simon Roburrin (LIGM) (author)
  • Updated on:

    June 29, 2021, 4:47 p.m.
  • Duration:

    00:22:08
  • Number of views:

    11
  • Type:

  • Main language:

    French
  • Audience:

    Other
  • Discipline(s):

 Downloads

 Embed/Share

Social Networks

 Options
Check the box to autoplay the video.
Check the box to loop the video.
Check the box to indicate the beginning of playing desired.
 Embed in a web page
 Share the link
qrcode