Loki: Adversarial Machine Learning for Network Security

Research Project, 2025 –

AI-driven network security systems are only as strong as their resistance to adversarial manipulation. Attackers who understand that a classifier or anomaly detector guards the network can craft inputs specifically designed to evade it, turning the intelligence of the defense into a liability. This project investigates adversarial machine learning in two critical and underexplored network security domains: BGP hijacking detection and encrypted traffic classification. The overarching goal is to understand the attack surface of ML-based detectors and develop principled defenses that remain robust under adaptive, intelligent adversaries.