Extraction and utilization of uncertainties in probabilistic deep neural net-works for object detection. Compared several state-of-the-art methods tocapturing uncertainties, and proposed novel ways of feeding uncertainty in-formation back into a network. Researched the effects of output distributionquality on the reliability of deep neural networks.

Worked on an automated inspection and maintenance system forurban drainage systems. Contributed to all parts of the pipeline,utilizing both traditional computer vision techniques for datapreparation and machine learning for automation. Worked as partof a team on production level neural networks, and led researchefforts in improving the reliability and interpretability of neuralnetwork architectures.Later on wrote my M.Sc. thesis as an industry thesis with Hades Technologies AG.