TDDE91 · 6 credits · Advanced level (A1N) · Informationsteknologi, Datateknik, Datavetenskap
Security and Privacy in Machine Learning
Active · next start HT 2027
No exam statistics yet
There is no exam sitting with at least 5 students in the last three academic years. New courses and project-only courses often have no statistics.
Part of 7 programmes
- Master of Science in Computer Science and Engineering
- Master of Science in Information Technology
- Master of Science in Computer Science and Software Engineering
- Master of Science in Engineering Mathematics
- Master of Science in Applied Physics and Electrical Engineering
- Master's Programme in Cybersecurity
- Master's Programme in Computer Science
About the course
Aim
The course aims to equip students with an understanding of security and privacy issues in modern machine learning systems and to introduce key concepts related to the trustworthiness of such systems. The course combines theoretical knowledge with practical skills for developing robust, privacy-preserving, and trustworthy machine learning systems. After completion of the course, the student shall be able to:
- Explain the major security and privacy risks in modern machine learning systems.
- Apply threat modeling and security analysis methods to identify and assess risks in machine learning pipelines, including evolving attacker models such as AI-assisted adversaries.
- Design and implement security and privacy attacks against machine learning systems, develop the corresponding defenses, and evaluate their effectiveness through empirical experiments.
- Explain core principles for building trustworthy machine learning systems and connect these concepts to real-world scenarios.
- Collaborate and communicate the results of a group project, including implementation, documentation, and demonstration.
Content
- Vulnerability analysis and threat modeling for machine learning systems deployed across various domains (e.g., healthcare, network management, and industrial environments) and different training approaches including supervised learning, reinforcement learning, distributed learning, and generative models, taking into account emerging attacker capabilities such as AI-assisted attacks.
- Security: Adversarial examples, data and model poisoning, backdoor attacks, and defenses against such attacks, including robust training and detection.
- Privacy: Model extraction, membership inference, leakage of sensitive training data, unauthorized use of models and datasets, secure model deployment, and privacypreserving technologies for machine learning.
- Concepts and terminology related to regulatory compliance, governance, reliability, fairness, and bias mitigation in modern AI applications.
- Scheduled hours
- 36 h
- Main field of study
- Informationsteknologi, Datateknik, Datavetenskap