Skip to content

Latest commit

 

History

History
20 lines (16 loc) · 739 Bytes

README.md

File metadata and controls

20 lines (16 loc) · 739 Bytes

VisReg

Implementation of the visualization regularizer (https://arxiv.org/abs/1604.02646) for neural networks trained on image tasks.

Instructions

Pre-requisites

The following python packages are required to be installed:

File organization

  • Data is located in cifar/data and mnist/data.
  • Logs generated by tensorflow summarywriter is stored in cifar/logs and mnist/logs.
  • Models are stored in cifar/models and mnist/models.

Training

The file mnist/train.py and cifar/train.py can be used to train the appropriate model by changing the parameters beta, gamma, theta for L2, VR2, VR1 regularizer coffecients.