Skip to content

agusgun/UnderstandingGroupNorm

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

27 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Exploratory Repository to Understand Group Normalization and Improving the Layer

Important

  • All of the VGG16 written in the notebook is VGG11. We are very sorry for these writing mistakes.
  • All of the VGG plotted in the notebook is also VGG11.

Understanding Part - Analysis Framework - How Does Batch Normalization Help Optimization

  1. Does a time-varying perturbation affect the training of GN?
  2. The impact of GN layer on the training process of neural network
  3. The impact of regularization on the training process of neural network with GN

Improvement Part

  1. Propose a new normalization layer by incorporating the underlying mechanism of BN to GN layer

Guide - Improvement Experiment

  1. Go to the main_training_script directory
  2. Run python assert_script.py to download all of the needed datasets and check
  3. Run python main.py with additional parameters based on your choice

Reference

  1. https://github.com/AlexeyGB/batch-norm-helps-optimization
  2. https://github.com/kuangliu/pytorch-cifar
  3. https://github.com/bearpaw/pytorch-classification