Christian Schroth, since April 2020 Research Associate with the Signal Processing Group, wins the Ernst-Blickle-Studienpreis 2020 for his Master’s Thesis on Robust Bayesian Cluster Analysis, which was supervised by Michael Muma. In his work, he dealt with robust cluster analysis and developed highly innovative algorithms of statistical learning to identify clusters within a given data set. To this end, the data are modelled as multivariate random variables, which form point clouds in a vector space, the so-called feature space. Those point clouds are to be explained by a suitable underlying statistical model. Both, the number of clusters, as well as the cluster memberships of the individual data points are unknown. A variety of established approaches to this problem exist, but they often lose their statistical guarantees or break down entirely in the presence of outliers or non-symmetrical spread around the cluster centers.
Christian Schroth’s work is remarkable, since he established a novel and coherent mathematical formulation for robust cluster analysis in a Bayesian framework, which allows to determine the a posteriori statistically most likely cluster model. The value of his theoretical work is underlined by applications to a set of simulated and real-data examples. For example, Christian’s method helped analysing a radar data set that was recorded previously by the Signal Processing Group during a study on radar-based gait analysis.
The SEW-Eurodrive Foundation was founded on November 30, 1989 and awards yearly prizes to student projects in the fields of electrical engineering, mechanical engineering and economics since 2002.