Themen

GNSS Spoofing Detection in V2X environments  – English only

Global Navigation Satellite Systems (GNSS) are a critical source of positioning information for modern vehicles and cooperative driving applications, yet they are vulnerable to spoofing attacks that can deliberately manipulate location estimates. This seminar investigates state-of-the-art approaches for GNSS spoofing detection. In addition, existing datasets for evaluating GNSS spoofing detection methods, as well as available open-source and research implementations, are systematically analyzed.

Artur Hermann

AI-Based Misbehavior Detection Systems in Vehicular Networks  – English only

Vehicular networks rely on cooperative message exchange to enable safety-critical and efficiency-enhancing driving functions, but are vulnerable to misbehavior caused by faulty or malicious participants. This seminar investigates artificial-intelligence-based misbehavior detection systems (MDSs), including machine-learning and deep-learning approaches for detecting manipulated or implausible V2X messages. In addition to detection methods, commonly used datasets, feature representations, and evaluation methodologies are systematically analyzed.

Artur Hermann

Post Quantum Crypto  – English only

Since the build of small quantum computers (QC) in combination with a quantum-algorithm which can solve the discrete logarithm problem very fast, there has been a growing need for new algorithms. The new algorithms should not be broken by a QC, but should still be feasible with our current CPUs. Your task is to take a dive into the topic, but you can choose the focus, e.g. an overview of current algorithms, a detailed explanation how one algorithm works, ...

Michael Wolf

Autonomous Driving Simulation with V2X Communcation  – English only

Autonomous driving simulation with vehicle-to-everything (V2X) communication research aims to create high-fidelity virtual environments in which connected vehicles interact with roadside equipment, cloud services, pedestrians, and other vehicles to simulate real-world traffic dynamics. Students can investigate how cooperative perception, intent sharing, and coordinated moves increase security, safety, and efficiency without putting physical assets at risk by infusing standardized V2X message sets into simulators like CARLA.

Dennis Eisermann

Adversarial Attacks against Autonomous Drone Systems  – English only

This seminar investigates how small, deliberate disturbances, known as adversarial attacks, might confuse autonomous drones' vision, navigation, and decision-making systems. We will review recent advances in developing such attacks against vision, LiDAR, and sensor-fusion pipelines, as well as discuss real-world case studies in which drone operations were compromised.

Dennis Eisermann

Privacy-Preserving Data Analysis with Differential Privacy  – English only

With the rapid growth of data-intensive technologies and large-scale analytics, concerns over individual privacy have become increasingly prominent when data is published. Differential privacy has been proposed as a formally sound solution to these concerns. In this seminar, you should introduce the fundamental concepts and mechanisms underlying differential privacy, examine its adoption in practical applications, and discuss its limitations.

Lukas Pietschmann

Expressing the "Right to Withdraw" in Policy Languages  – English only

Psychology research relies on shared data repositories to promote transparency and reproducibility. However, adhering to data protection regulations requires dynamic consent revocation. When a participant chooses to opt out of a study after their data has already been stored, it needs to be deleted retroactively. This seminar paper will conduct a comparative literature analysis of Privacy Policy Languages to determine which is best suited for these revocation workflows.

Lukas Pietschmann

Cross-User collaborative behavioral anomaly detection in zero trust networks  – English only

This seminar explores approaches for detecting anomalous behavior across collaborating users in zero trust networks. Rather than applying anomaly detection on a single user in isolation, this seminar extends the view to collaboratively working teams where an attacker moves slowly and therefore would normally stay below a certain threshold to remain undetected.

Janek Schoffit

Incremental Policy Deployment for Microsegmented Zero‑Trust Networks  – English only

This seminar explores techniques for efficiently updating access policies in microsegmented Zero Trust Architectures by compiling and applying only the minimal set of changes required. It focuses on how existing policy deployment can be improved to minimize update cycles and enhance adaptability in dynamic network environments.

Janek Schoffit

LLMs - friend or foe?  – English only

Since the appearance of the 2022's consumer-facing Large Language Model (LLM) chatbot, ChatGPT, LLMs have gained a vast user base and have been exploited in a variety of different disciplines, from law to science. This paper aim to i) explain the mechanics of the inner-workings of LLMs, ii) explore the successful use-cases of LLMs in different disciplines where they have brought great value, while at the same time iii) reflecting on the limitations of LLMs, such as hallucinations, biases, limited reasoning, etc.

Natasa Trkulja

Automatic Software Vulnerability Detection and Patching with Large Langauge Models  – English only

Software vulnerabilities pose significant risks to organizations, as they can enable attackers to compromise critical IT infrastructure and sensitive data. This seminar explores the state of the art in LLM-based software vulnerability detection and automated patching, covering techniques, systems, datasets, and real-world applications. In addition, we identify key research gaps and discuss directions for addressing them.

Jonas Weßner

Network Vulnerability Detection and Mitigation with LLMs  – English only

Software vulnerabilities are a persistent risk in IT infrastructures, as they can allow attackers to compromise systems and data. To mitigate these risks and enforce least-privilege access, a wide range of access control mechanisms, such as router and host firewalls, authentication, and two-factor authentication, are employed. When new vulnerabilities are discovered, these mechanisms can be leveraged to isolate high-risk devices and services. This seminar reviews the state of the art in research on automatically detecting and mitigating software vulnerabilities in computer networks through LLM-driven automated access policy updates. We also identify open research gaps and outline directions for future work.

Jonas Weßner

Comparison of Speculative Consensus Protocols  – English only

Consensus protocols are important for modern-day distributed systems. They are, for example, used to order input for distributed databases or state-machine replicated systems. Running a full consensus protocol, however, can be quite expensive and lead to higher latencies. Speculative consensus protocols aim to enhance the performance of these systems by optimistically assuming fault-free execution in the common case, while still ensuring correctness in the event of faults. This seminar should examine different speculative BFT consensus protocols and provide a comparison of them. The student can decide the focus of the seminar by either comparing more protocols on a higher level or an in-depth discussion of only a few.

Christoph Denzel

Asynchronous BFT Consensus Protocols  – English only

The FLP impossibility theorem states that it is impossible to deterministically reach consensus in an asynchronous network if there is just a single faulty node. Up until recently, the majority of practical BFT consensus protocols evaded the FLP impossibility by making network timing assumptions. As a consequence, these consensus protocols can only guarantee liveness if the underlying network meets these assumptions. In recent years, a new family of BFT consensus protocols has emerged, which builds on probabilistic algorithms and is able to ensure liveness in fully asynchronous networks with high probability. These BFT consensus protocols are mostly adopted in blockchain systems, because their construction allows to tolerate network partitions of arbitrary length. The goal of this seminar is to dive into the internals of two or three representative examples for asynchronous BFT protocols, such as HoneyBadger, DAG-Rider or Bullshark, and to elaborate their core concepts.

Alexander Heß

Consensus algorithms using trusted components  – English only

To keep replicas of a fault-tolerant service consistent, consensus protocols guarantee that the order of incoming requests is the same in each replica even though some of the replicas have crashed or exposed arbitrary misbehaviour. Such protocols need at least 3f+1 replicas in order to mask f faulty ones. However, with the use of trusted components that are not supposed to fail, the number can be reduced to 2f+1 which saves resources and reduces latencies. For this topic, the student is supposed to give a review of a couple of approaches using trusted components. Their principle internal procedures should be presented and the approaches should be compared against each other.

Franz J. Hauck

Software Transactional Memory: Mechanisms for atomicity and isolation  – English only

This seminar topic investigates how Software Transactional Memory (STM) systems implement atomicity and isolation in shared-memory programs. The student will study and explain the core mechanisms used in modern STM designs, including read and write set tracking, versioning, validation, commit protocols, and rollback. The seminar should compare alternative design choices (e.g., eager vs. lazy versioning and validation) and explain their correctness implications. The goal is to build a clear mental model of how STM systems work internally.

Franz J. Hauck