The Machine Learning (MLE) group is concerned with the development of novel machine learning methods with a particular focus on deep learning. These methods are detached from concrete applications, but are oriented towards the needs of the DLR. Therefore, the group works at the interface between basic research in Machine Learning and concrete tasks at the DLR. For example, the group innovates in the areas of architecture design, training methods or scaling and implements these in prototypes.
The main focus of the MLE group is on the topics of uncertainty, anomaly detection and explainability. In the area of uncertainty, the aim is to add uncertainty estimates to the predictions of deep neural networks in order to be able to evaluate the reliability of these predictions and to use or develop more robust methods. In the topic of anomaly detection, the group investigates and develops machine learning methods, which, for instance, allow to find time points or intervals in sequential data that deviate from the normal state. The topic of explainability deals with methods that make decisions of deep neural networks more understandable and interpretable with respect to their predictions, even for laymen, in order to promote their use also outside the academic environment.
Further projects of MLE group are concerned with the integration of prior physical knowledge into statistical models, the investigation of erroneous training data, and the application of machine learning techniques in quantum imaging.
In addition, the group is available to consult internal and external partners and participates in the networking of the regional machine learning community.