2023-06-252023-06-25https://tore.tuhh.de/handle/11420/16747The research area of Machine Learning provides methods and data-driven statistical models that "learn" through experimentation or observation. This is advantageous for complex problems where it becomes very difficult if not impossible to design exact models. As a consequence, deep learning achieves breakthroughs mainly in areas where models do not exist at all or human-made models capture only small parts of the reality. The highly dynamic problem space of aeronautical communications in a shared radio spectrum is such a field. While Machine Learning-based methods promise good performance, the common pitfall is that such algorithms provide a well-performing blackbox. In aeronautical communications, however, strict performance guarantees have to be made and kept. In this project, the Institute of Communications and the Institute of Communication Networks will join forces to turn Machine Learning-based methods into a scientifically rigorous approach. Machine Learning-based methods are to be applied, evaluated and explained to problems in the field of communications engineering. These shall be shown to work within strict performance bounds required for safety-critical applications. For this, expert domain knowledge is to be combined with data-driven learning algorithms. The understanding of internal processes is focused upon, and the information bottleneck method shall be applied to gain understanding of the intricacies of deep learning methods. At the same time, novel efficient signal processing units shall be designed, leveraging the information bottleneck method, to develop deep learning solutions that are practically implementable in hardware-constrained communication devices. The ongoing cooperation with scientific and industrial partners as well as the preparation of a project application at the DFG are key objectives of this endeavor.