This repository contains the bachelor thesis "Attacks against Deep Learning based Network Intrusion Detection Systems" by Dominik Brockmann. The work evaluates the vulnerability of deep learning-based network intrusion detection systems to adversarial attacks, assessing the impact of different datasets, model architectures, and attack methods.
Cyber-attacks continue to pose serious risks to information systems, and while deep learning improves intrusion detection through better generalization, it also introduces a vulnerability: adversarial attacks that exploit minimal changes in input data to cause unexpected outputs. This thesis investigates how white-box, substitute, and black-box attack methods succeed against deep learning-based NIDSs, indicating that the underlying data characteristics contribute more to these vulnerabilities than the model architectures themselves.
This thesis has been further developed and refined. The updated code and extended research can be found in the repository: now-you-see-me-now-you-dont.