Prof. Dr.-Ing. Bodo Rosenhahn

Our group focuses on the one hand on the field of photonic integrated circuits (PICs) through novel inverse design methods and optimization techniques.

A primary focus of our research is the development of inverse design techniques for polymer-based optical components in PICs. We apply our experience with large-scale machine learning to create designs under complex constraints. To address the computational challenges of this process, we employ techniques from our research on video compression to reduce the memory requirements of optimizing polymer-based PICs with a large spatial footprint at high resolution. We are also adapting the inverse design process to account for the fabrication of the structures which we observe with in-situ microscopy.

The second field of research is located in the domain of out-of-distribution (ood) or anomaly detection. Here we make use of normalizing flows to map a distribution of data samples to a normal distribution and use statistical measures to provide scores for anomalies. As the used neural networks are invertible, they also allow a reasoning of the parts which cause the ood score.