Research Group
Machine Learning
and Security
and Security
View from our building over Berlin.
We offer a number of courses each semester that revolve around machine learning and security. These include lectures on learning algorithms in security systems and adversarial machine learning as well as our labs where people can experiment with attacks and malicious code. Teaching is fun for us and so we have been able to even win awards for our lectures and practical courses.
AML — Adversarial Machine Learning
This integrated lecture is concerned with adversarial machine learning. It explores various attacks on learning algorithms, including white-box and black-box adversarial examples, poisoning, backdoors, membership inference, and model extraction. It also examines the security and privacy implications of these attacks and discusses defensive strategies, ranging from threat modeling to integrated countermeasures.
This lab is a hands-on course that explores machine learning in computer security. Students design and develop intelligent systems for security problems such as attack detection, malware clustering, and vulnerability discovery. The developed systems are trained and evaluated on real-world data, providing insight into their strengths and weaknesses in practice. The lab is a continuation of the lecture "Machine Learning for Computer Security" and thus knowledge from that course is expected.
WORMS — LLM-based Computer Worms
This project explores LLM-based computer worms and the potential of large language models to enable self-replicating malware. In a strictly isolated virtual environment, participants will develop and evaluate proof-of-concept worms that use LLMs to explore their environment and propagate across simulated machines. The project will also investigate the capabilities, limitations, and defense mechanisms for such threats.
RAID — Reproducing AI Attacks and Defense
This project puts recent AI research to the test. Participants will re-implement current attack and defense techniques that utilize machine learning, evaluate their capabilities, and design improvements. Possible techniques include attacks and defenses for large language models and computer vision systems. The overall goal is to learn about the state of the art in AI security and reproduce results where possible.
ALMO — Attacks on Large Language Models
This block seminar focuses on attacks against large language models. We will examine recent attacks such as jailbreaks, indirect prompt injection, and data poisoning, and discuss their impact on practical security and privacy. We will also look at possible defenses and countermeasures along with their limitations. The seminar is intended for Master students with a background in machine learning.
CARE — Code Analysis and Reverse Engineering
This block seminar is concerned with the analysis and reverse engineering of code. We will cover different techniques for program analysis of source code and binary code. In addition, we will look at concepts for understanding unknown software, reverse engineering its functionality, and discovering security vulnerabilities. The seminar is intended for Master students.
Below is a list of all the courses we have offered in recent years. Note that some courses are not offered regularly, while others are planned and not yet available. Please consult the respective pages on the ISIS platform of TU Berlin.
Are you looking for an exciting topic for your Bachelor or Master thesis? We offer research-oriented thesis topics at the intersection of machine learning and computer security. The full list of topics is available exclusively through the STROD portal of TU Berlin.
As we have only a limited number of thesis slots, we require successful participation in relevant courses to ensure a good match. Please read the topic descriptions and requirements carefully. If you have any questions, feel free to contact the supervisors listed for each topic.
