Themen

✘ Recent Advances in Autonomous Driving – English only

Autonomous Driving as well as other AI-based applications have the potential to shape our future. The development of advanced and efficient machine learning algorithms made Autonomous Driving possible nowadays. In this seminar, you will address autonomous driving from both points of view: the theoretical (based on the literature) as well as the practical one. In particular, you will survey and compare the existing work in this field and try to answer the question which machine learning technique performs best in Autonomous Driving. Moreover, you need to address (based on the literature) the question of how V2X communication can enhance the autonomous driving experience.

Ala'a Al-Momani

✘ Deep Learning vs. Reinforcement Learning: When to Use Which? – English only

Deep Learning (DL) allowed Google to master the game of Go a decade earlier than expected through building an AI machine that is capable of beating expert players at the mentioned game. On the other hand, Reinforcement Learning (RL) allows, experimentally, AI machines to figure out things that no programmer could teach them and it is considered one of the top ten breakthrough technologies in 2017 according to the MIT technology review. In this seminar, you will investigate both learning techniques while addressing what benefits each one brings over the other. In addition, in what applications each one fits the best. Moreover, you need to address the best practices in these emerging technologies. By this, you also need to investigate real-world DL- and RL-based applications and try to figure out why this has been used and not the other. Previous knowledge in ML/DL/RL/NN/AI is preferred and it will help you progress faster in this seminar.

Ala'a Al-Momani

✘ Architectures for Data-intensive Applications – English only

Data-intensive applications require reliability, scalability, and maintainability. This seminar topic addresses system architectures that enable such applications and introduces the relevant concepts and mechanisms.

Benjamin Erb

✘ TensorFlow – English only

Developed by Google TensorFlow is an open source software library for machine learning. It uses data flow graphs for numerical computation. The flexible architecture allows to deploy computation to one or more CPUs or GPUs in a desktop, server, or mobile device with a single API. The goal of this seminar is to give introduction to TensorFlow, its usecases and deeper characteristics.

Note: basic knowledge in Python is recommended.

Eugen Frasch

✘ Modern Encryption for File Sharing  – English only

In modern applications we value confidentiality of files we sent (e.g. pictures, bank statements, contracts). Simple encryption will force us to share the password with everyone. Some systems will allow us to specify possible recipients with public keys. The goal of this seminar is to research these systems and provide an introduction to a concrete and secure shareable encryption scheme.

David Mödinger

✘ Misbehavior Detection in Vehicular Networks – English only

One approach to improve road safety and efficiency in the future is by allowing vehicles to communicate and warn or plan their routes. This is realized through vehicular ad-hoc networks. These networks improve safety and efficiency by sending messages to, for example, inform vehicles in a particular area that a traffic jam is ahead of them. For the correct and safe operation of applications, it is essential that the information sent by other vehicles can be trusted or verified. In this seminar paper, you will discuss one or two new approaches to detect misbehavior of vehicles in these networks, and put these into the context of existing work.

Rens van der Heijden

✘ Alignment-free Protocol Message Format Reverse Engineering – English only

Understanding the communication of networked systems without knowing the protocol specification can be achieved by the reverse engineering of the network traffic. Examples for use cases of this approach are the analysis of botnets and vulnerability detection in network protocol implementations. Several methods to infer the protocol message format use the Needleman-Wunsch-algorithm-based progressive multiple sequence alignment. Instead, also alignment-free methods to find structure in byte sequences are available. An especially interesting example is the Leimeister et al. alignment-free method.

The goal of this seminar topic is to identify commonalities and differences between alignment and alignment-free algorithms with focus on applicability to infer message formats of protocols and to discuss their respective strengths and weaknesses.

Stephan Kleber

✘  Cryptography Engineering – English only

Every day we rely on the security of applied cryptography (e.g. security protocols) to protect our data. This application of cryptography in real world solutions is difficult and prone to errors; one mistake can compromise the entire security.
Cryptography engineering is the discipline of creating these real world solutions and entails two different aspects: the creation of cryptographic protocols by combining several cryptographic primitives (e.g. MAC and symmetric encryption), and the implementation (software / hardware) of cryptographic protocols.
The goal of this seminar paper is to give an overview of cryptography engineering with a focus on dos and don'ts, and best practices for creating and implementing cryptographic protocols.

Note: It is highly recommended that the student has basic knowledge of cryptography, for example by having attended a cryptography and/or IT security lecture.

Dominik Lang