https://github.com/yeonghyeon/adae-tf

TensorFlow implementation of Anomaly Detection with Adversarial Dual Autoencoders (ADAE) with MNIST dataset.
https://github.com/yeonghyeon/adae-tf

adversarial-dual-autoencoders anomaly-detection mnist mnist-dataset tensorflow

Last synced: 8 months ago
JSON representation

TensorFlow implementation of Anomaly Detection with Adversarial Dual Autoencoders (ADAE) with MNIST dataset.

• Host: GitHub
• URL: https://github.com/yeonghyeon/adae-tf
• Owner: YeongHyeon
• License: mit
• Created: 2021-03-18T02:44:43.000Z (over 4 years ago)
• Default Branch: master
• Last Pushed: 2021-03-26T09:13:07.000Z (over 4 years ago)
• Last Synced: 2025-01-08T18:54:24.180Z (9 months ago)
• Topics: adversarial-dual-autoencoders, anomaly-detection, mnist, mnist-dataset, tensorflow
• Language: Python
• Homepage:
• Size: 1.84 MB
• Stars: 4
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
[TensorFlow] Anomaly Detection with Adversarial Dual Autoencoders

=====

TensorFlow implementation of Anomaly Detection with Adversarial Dual Autoencoders (ADAE) with MNIST dataset.  

The Keras implementation is provided as the following link.  

https://github.com/kjm1559/ADAE_LSTM_Autoencoder

## Architecture

### Objective Functions



    

  
The objective functions (losses) for training ADAE [1].



### ADAE architecture



    

  
The architecture of ADAE.



### Graph in TensorBoard



    

  
Graph of ADAE.



### Problem Definition



    

  
'Class-1' is defined as normal and the others are defined as abnormal.



## Results

### Training Procedure



  


    

    

    

  

  
Loss graphs in the training procedure.Each graph shows the generative loss, and the two terms that make loss-G.





  


    

    

    

  

  
Loss graphs in the training procedure.Each graph shows the discriminative loss, and the two terms that make loss-G.





    

  
Restoration result by ADAE.



### Test Procedure



  

  
Box plot with encoding loss of test procedure.





  


    

    

    

  

  
Normal samples classified as normal.


  


    

    

    

  

  
Abnormal samples classified as normal.


  


    

    

    

  

  
Normal samples classified as abnormal.


  


    

    

    

  

  
Abnormal samples classified as abnormal.



## Environment

* Python 3.7.4  

* Tensorflow 1.14.0  

* Numpy 1.17.1  

* Matplotlib 3.1.1  

* Scikit Learn (sklearn) 0.21.3  

## Reference

[1] Ha Son Vu et al. (2019). Anomaly Detection with Adversarial Dual Autoencoders. arXiv preprint arXiv:1902.06924.