KDD-OpenSource / QUGA

This repository contains both supplementary material and the code for the publication "QUGA Quality Guarantees for Autoencoders Using Unsupervised Adversarial Examples"

QUGA: Quality Guarantees of Autoencoders via Unsupervised Adversarial Attacks.

This repository contains Code and supplementary material for the ECMLPKDD2020 Paper Quality Guarantees for Autoencoders via Unsupervised Adversarial Attacks.

Usage

git clone git://github.com/KDD-OpenSource/QUGA.git  
virtualenv venv -p /usr/bin/python3  
source venv/bin/activate  
pip install -r requirements.txt  
python3 main.py

Reproduction of Experiments:

To reproduce the results of the sine-curve dataset replace the contents of objectCreator.py by the contents in FullSynthetic.py. Then run "python3 main.py". Similarly to reproduce the experiments with the ECG5000 Dataset replace the contents of objectCreator.py by the contents in FullReal.py.

Supplementary Material

The file "QUGA_withSupplMat.pdf" contains both the paper and supplementary material (proofs of the theorems and a precise description of the experiments).

Authors/Contributors

• Benedikt Böing
• Rajarshi Roy
• Emmanuel Müller
• Daniel Neider