Bachelor and Master Theses

This is also for M.Sc in Cybersecurity 

As ML models increasingly support high-risk domains (healthcare, autonomous systems, cybersecurity), they remain vulnerable to adversarial attacks—subtle perturbations that cause misclassification while appearing normal to humans. Most defenses are tailored to synthetic attacks or toy datasets and fail under real-world conditions (transfer attacks, physical-world attacks, distribution shifts).

There is a strong need for robust, explainable adversarial defenses suitable for applied industry settings.

Problem Statement

How can we design practical adversarial defenses that remain robust under realistic threat models while providing transparency for decision-makers?

ML/DL Techniques

• Random Forest / Gradient Boosting
• Autoencoders for anomaly detection (zero-day threats)
• 1D CNN / LSTM / Transformer-based IDS models

Tools

• Python, PyTorch/TF
• Wireshark/Tshark for traffic analysis
• Splunk/ELK for SOC simulation (optional)