About me

I am a PhD student at the chair of mathematics of information processing at LMU Munich with a strong interest in the intersection of mathematics, signal processing, and deep learning. My research focuses on applications of high-dimensional probability theory, with current work spanning two distinct areas.
My PhD journey began with high-dimensional covariance estimation, inspired by challenges in modern wireless communication systems with massive multiple-input multiple-output (MIMO) antenna arrays. Recently, I have expanded my focus to explore deep learning approaches for crack detection in digital images of oil paintings. This project allows me to combine my technical background with a long-standing passion for the connection between mathematics and the arts.

Research interests
High-dimensional probability theory | Covariance matrix estimation | Mathematics & Arts

Short bio

I hold a Bachelor's degree in Mathematics from the University of Bremen, where I chose "Art, Media, and Aesthetic Education" as applied subject. I went on to complete a Master's degree in Mathematics at Friedrich Schiller University Jena, where I wrote my thesis on biorthogonal wavelet bases with Prof. Dr. Winfried Sickel. I began my PhD studies at RWTH Aachen and am continuing them at LMU Munich under the supervision of Prof. Dr. Holger Rauhut.

Research topics

Covariance Matrix Estimation for Massive MIMO

Massive multiple-input multiple-output (MIMO) communication systems are a key technology for modern wireless communication and play a central role in the development of fifth-generation (5G) cellular networks. In massive MIMO, base stations are equipped with a large number of antennas, offering a high degree of spatial freedom and enabling simultaneous communication with multiple user terminals. Due to the typically limited angular spread in such systems, the channel vectors corresponding to individual users tend to lie in low-dimensional subspaces. Our goal is to identify, for each user, a low-dimensional beamforming subspace that captures most of the signal power. This signal subspace estimation problem can be tackled by estimating the user signal covariance matrix. To this end, we investigate estimation error bound in a truncated version of the nucelar norm for the sample covariance matrix as estimator. The bound should scale in the number of observed time samples, the number of sampled entries (antennas), the truncation and noise level instead of the input dimension.

This is joint ongoing work with Sjoerd Dirksen from Utrecht University and my supervisor Holger Rauhut from LMU Munich.

Deep learning aided crack detection in digitized oil paintings

Advancements in imaging technologies - such as high-resolution photography, infrared reflectography, and X-ray radiography - combined with recent breakthroughs in deep learning, have opened up a wide range of opportunities for the investigation and conservation of artworks. These methods allow for the non-invasive analysis of the internal and surface structure of a painting offering valuable insights into its material composition and condition. A particularly important application of these techniques lies in the automated detection of cracks in digitized paintings. Cracks are among the most common forms of degradation found in aged artworks and carry significant information about the painting's age, environmental history, and structural stability. Detecting and analyzing them accurately is essential for conservators in planning and executing restoration measures. Motivated by this, in this project we explore deep learning-based methods for reliable crack detection in high-resolution digital images of paintings. The goal is to develop models that can distinguish cracks from visually similar elements such as brush strokes and that generalize well across different styles, painting techniques, and image modalities.

Beyond practical conservation use, this research topic also contributes to the broader interdisciplinary field connecting mathematics, art history, and cultural heritage research. It exemplifies how modern mathematical and algorithmic tools can support the preservation of historical and cultural artifacts, highlighting the enriching potential of collaborations between technology and the arts.

This is joint ongoing work with my supervisor Holger Rauhut from LMU Munich.