Systems Security and Machine Learning (212071-WS 26/27)

Machine learning is transforming both the methodology and the attack surface of systems security: it serves as a tool to help us solve security challenges, but also represents a new class of targets for systems security. The course covers both aspects and focuses on the following topics: - The course introduces the ML concepts needed for security reasoning, including datasets, evaluation metrics, embeddings, transformers, adversarial behavior, and similar topics. - ML for systems security: on the one hand, we study how ML supports software testing, vulnerability detection, reverse engineering, binary analysis, malware analysis, and similar topics in systems security - Security of ML-enabled systems: on the other hand, we also cover topics such as adversarial ML, prompt injection, LLM application security, insecure code generation, and the security risks of autonomous coding agents. - A recurring theme is how to critically assess ML/security claims, including threat models, dataset quality, label noise, benchmark leakage, baselines, and deployment realism.