On Training and Verifying Robust Autoencoders

This repository contains the code and experimental configuration files for the DSAA2022 Paper 'On Training and Verifying Robust Autoencoders'

Usage

Please note that in order to make use of our solution framework you need to install Marabou

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

Reproduction of Experiments:

To reproduce the results of the any experiment in the paper first run

python3 main.py configs/reprod/dsaa/config_train_EXPERIMENT.cfg

for your choice of experiment. This creates the necessary models. Thereafter store the model in e.g. 'models/reprod/autoencoder_EXPERIMENT' and run

python3 main.py configs/reprod/dsaa/config_test_EXPERIMENT.cfg

Make sure that the path in the respective test configuration file poins to where you have stored the previously trained models. You will find the results in the folder 'reports'.

Authors/Contributors

• Benedikt Böing
• Emmanuel Müller