Convolutional Auto-encoders

This repo is aiming to provide a set of Convolutional Auto-encoder implementations (CAEs) for Deep Learning.

Convolutional Auto-encoder did not draw so much attention since it's proposed. However, I think that this model offered a very nice unsupervised feature learning model from neural network persepective.

Following dependencies are required:

• Python 2.7.8
• numpy
• scipy
• theano

You can also use Anaconda directly, this python distribution will offer you all dependencies.

##Updates

• ConvNet layer [20141021]
• Original ConvNet Auto-encoder [20141021]
• Tested for AWS GPU instance [20141025]
• Example for classification [20141025]
• Add some support functions for ConvNet Layer [20141027]

##To-do

• Multiple activation function support
• Support functions for ConvNet Layer and ConvNet AE
• Sparse ConvNet Auto-encoder
• Stacked ConvNet Auto-encoder

##Notes

• All experiment in this repo are conducted on GPU, in order to run it faster, you are suggested to have a GPU on your machine.

• If you forked this repo, use and modifiy update.sh to avoid updating unnecessary files and data.

• Tested on AWS GPU instance, the performance looks like Tesla K20C.

##Contacts

Hu Yuhuang
Advanced Robotic Lab
Department of Artificial Intelligence
Faculty of Computer Science & IT
University of Malaya
Email: [email protected]