The aim of this project is to start off new collaborations between computer science and applied mathematics within the domains of deep learning and stochastic nonlinear optimization. Our central goal is to explore novel avenues to apply deep learning to inverse problems by bringing together our existing expertise in the respective fields, and increase efficiency of existing methods using modern optimization approaches. We thus build up core methodology which will ultimately be useful for different applications within other projects of the cluster, in particular efficient low-level imaging and image processing systems in the imaging hangar. While we kick off our collaboration with a single postdoc and at first limited scope to assess viability of the approaches, it is planned to branch into other aspects of optimizing deep models in later projects where advanced mathematical methods can also lead to novel insights and improvements.
Back to current projects.