In collaborations with key academic partners at home and elsewhere, we have developed custom solutions for inference problems in the sciences. Here are some examples.
In a DFG funded collaboration with Mark Bangert's group at the German Cancer Research Center (DKFZ) in Heidelberg, we helped develop efficient tools for the computation of uncertainty measures for radiation treatment planning systems. These can help reduce the risk of complications in tumor therapy. [1, 2, 3, 4, 5]
Constructing physical prototypes and conducting technical experiments is an expensive process. In a collaboration with Sebastian Trimpe's group, and colleagues at the ETH and CMU, our entropy search framework has engendered a principled framework for such multi-modal experimental design tasks. [1, 2]
Serial Block-Face Scanning Electron Microscopy provides three-dimensional images of neural tissue at unrivaled, full-volume detail. Extracting the connectome, the graph of neural connections, from these images is extremely challenging. We were fortunate to be able to provide a modest contribution to the impressive efforts of Moritz Helmstaedter's team to develop a comprehensive software solution to this daunting scientific task. 
The deposition and transport of carbon and other matter in the soil is an important modeling variable for climate science and other geoscientific questions. In an intramural collaboration with Thomas Scholten's group, funded by the Excellence Cluster for Machine Learning in the Sciences, we are currently building new machine learning models customized to efficiently and faithfully model this process.