The Machine Learning (MLE) group is focused on the development of novel methods of machine learning, in particular deep learning. These methods are detached from the concrete applications, but are oriented towards the needs of DLR. The group thus forms an interface between basic research in the field of machine learning and the tasks at DLR. For example, the group innovates in the areas of architectural design, training methods and scaling and converts these into prototypes.
Another focus is quality assurance for machine learning methods, for example with regard to uncertainty estimation, traceability and interpretability.
The group is also available to advise internal and external partners and is involved in networking the regional machine learning community.
Concrete topics include, for example:
- Machine Learning for high-resolution geodata
- Architectures for data fusion
- Scaling Deep Neural Networks to Large-Scale Problems
- Methods for small and incorrect data records
- Image and text mining from social networks
- Automatic architectural design
- Introduction of physical prior knowledge into deep learning approaches
- Combination of Sparse and Low Rank Methods with Deep Learning
The group will be established in close cooperation with Prof. X. Zhu of the Institute for Remote Sensing Methodology in Oberpfaffenhofen. It already cooperates with the chairs for Computer Vision (Prof. J. Denzler) and Theoretical Computer Science II (Prof. J. Giesen) of the FSU Jena.