AI for Cybersecurity: from Adversarial Anomaly Detection to Intelligent Network Security Systems

The speed in which attacks appear and propagate requires efficient cyber protections. New attacks are mounted with the support of modern Machine Learning (ML) and AI algorithms, e.g., by learning users’ behavior from data. Fighting such attacks requires systems that can learn attack patterns efficiently and autonomously. ML and AI represent an opportunity to evolve cyber-defenses, too. Here we address and summarize our research efforts in this direction: (1) anomaly detection in cybersecurity using ML tools and in presence of adversaries, (2) classification and generation of strings that are relevant for cybersecurity, in particular passwords, subdomains and phishing URLs, and (3) AI-powered network security, including new representations for darknet traffic and adaptive honeypots.