This repository contains the code and experimental configuration files for the ICDM2022 Paper Post-Robustifying Deep Anomaly Detection Ensembles by Model Selection

"Model_training" contains the code for generating the anomaly detectors used for this Paper. Also take a look at DEAN here: https://github.com/psorus/DEAN

"Model_verification" contains the code for verifying the given models. Note that in order to verify, one must have Marabou installed.

To verify a given model, first save in it e.g. './models/trained_models/deepsvdd_cardio/models' as demonstrated by the example. Thereafter you can run

python3 main.py configs/reprod/icdm/config_test_svdd_cardio.yaml

and find the results in './reports'. If you choose to save the model in a different folder you will need to adjust the config file accordingly.

Authors/Contributors

• Benedikt Böing
• Simon Klüttermann
• Emmanuel Müller