Skip to content

Implementation and improvement of paper 'Learning Multiple Views with Orthogonal Denoising Autoencoders'

License

Notifications You must be signed in to change notification settings

tengerye/orthogonal-denoising-autoencoder

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

19 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Orthogonal Denoising Autoencoders

Orthogonal denoising autoencoders are for multi-view learning problem (a.k.a. multimodal, model fusion). Data samples are collected from diverse sensors or described by various features and inherently have multiple disjoint feature sets.

The project is extensive light-weight, which is an implementation of the MMM2016 best paper.

Getting started

Each package is an implementation using different languages (frameworks). Therefore, you can use codes within a package independently and install corresponding dependent libraries.

Contents

  • python: implementation uses numpy and scipy to implement feed-forward, back-propogation and stochastic gradient descent.
    • orthdAE:implementation BLABLABLA.
  • tensorflow: implementation is build on the TensorFlow. So the details mentioned above are controlled by TensorFlow.

Dependencis

  • python: 2.7.13
  • TensorFlow: 1.3.0
  • numpy: 1.11.3
  • scipy: 0.18.1

Notice

  • The size of input for Python code is: NUM_OF_FEATURE x NUM_OF_EXAMPLE. The size of placeholder for TensorFlow code is: NUM_OF_EXAMPLE x NUM_OF_FEATURE.

Demo

Each package contains a demo (demo.py) which implements the first experiment of the MMM2016 best paper.

In order to run the demo, go to the corresponding directory and run the demos in shell:

python demo.py

You will see 3 figures pop up and then the terminal will print training process. In the end, the result of our model will pop up as figure.

Python implementation

Finish.

TensorFlow implementation

Finish.

Pytorch implementation

In progress.


Acknowledgments

Orthogonal denoising autoencodrs are proposed in the paper, Learning Multiple Views with Orthogonal Denoising Autoencoders, which won the best paper award in MMM2016. Certificate

If you have an question or suggestion, please contact the author: TengQi Ye(yetengqi@gmail.com). Your citation will be my biggest inspiration.

BibTeX Style Citation

@inproceedings{ye2016learning,
  title={Learning multiple views with orthogonal denoising autoencoders},
  author={Ye, TengQi and Wang, Tianchun and McGuinness, Kevin and Guo, Yu and Gurrin, Cathal},
  booktitle={International Conference on Multimedia Modeling},
  pages={313--324},
  year={2016},
  organization={Springer}
}

About

Implementation and improvement of paper 'Learning Multiple Views with Orthogonal Denoising Autoencoders'

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages