Practical course on Machine Learning Security

InfSec-Logo
ProfessorProf. Dr. Ghassan Karame
Course number212427
CP4
SWS2
MoodleLink

Dates

Practical sessions Thursdays, 14:30-16:00, Room: MC 1.84
First session 10.10.2024

Open for
This course is open to students of the Bachelor Program in IT Security, the Master’s program in IT Security / Information Engineering, IT Security / Networks and Systems, and the Master’s degree in Computer Science.

Content

In recent years, machine learning has been widely adopted and experienced a great surge with recent advances in machine-learning-as-a-service (MLaaS) platforms, large language models, such as ChatGPT, and generative AI, such as Stable Diffusion. We have reached a point where these technologies are so accessible that they found their way to consumer devices, yet the underlying function and problems are mostly opaque.

Often invisible to the user, these systems consist of sophisticated pipelines for data pre-processing and training before a model can be deployed for inference in practice. These pipelines often try to ensure the robustness of the underlying machine learning model, i.e., correct prediction of unseen inputs, yet they can easily fall victim to an attacker.

In this lab course, the students will develop an understanding for the fundamentals of machine learning pipelines, and their susceptibility for attacks by implementing concepts from state-of-the-art research papers. These fundamentals are further solidified by introducing concrete real-world use-cases, e.g., federated learning for attacks during the training phase or adversarial patches for attacks on deployed models.

Since the course offers close guidance, the number of participants is limited to 8. Interested students must enroll in the Moodle course before the first session on 10.10.2024, i.e., on 09.10.2024 at the latest. There will be a session each Thursday from 2:30 p.m. to 4 p.m. (except 31.10.2024). 

 

Requirements

Email application with motivational letter is required. We strongly recommend prior participation in one of the artificial intelligence classes at RUB, e.g., Introduction to Artificial Intelligence, Deep Learning, Machine Learning: Supervised Methods, or similar. Further, programming experience in Python is strongly recommended.

Language

English

 

Form of Examination

Practical assignments and presentations throughout the course.