Training VAE and an Mixed-Loss VAE with normal network traffic, then implementing validation and test steps that fold in the attack traffic to ensure models' reconstruction error satisfactorily separates data at with accuracy. Leveraging Bayesian optimization to tune hyperparameters and testing one anomaly at a time hold out strategy with AUCs.
python internet-of-things cybersecurity bayesian-optimization anomaly-detection variational-autoencoder network-trace roc-auc mixed-loss-variational-autoencoder
-
Updated
Jun 14, 2024 - Jupyter Notebook