Physics · Machine learning · Dynamical systems
I’m a physics PhD student at Heidelberg University, working on how machine learning can help us understand and predict dynamical systems.
Research
My research sits at the intersection of computational physics, machine learning and dynamical systems.
I work on reconstructing dynamical systems from time series and predicting their behaviour beyond the conditions seen during training. A particular focus is on non-autonomous systems, where the conditions under which they evolve change over time. I’m also interested in how models can generalize to new dynamical regimes, including across tipping points.
My approach combines mathematical analysis with data-driven modelling: understanding where existing models reach their limits, and developing methods that can go beyond them.
Central Institute of Mental Health, Mannheim
Heidelberg University
Publications & academic work
How well do learned models predict a system’s behaviour beyond their training conditions? We analyse structural limitations of existing reconstruction models and show how feature splitting enables predictions in unseen dynamical regimes, including across tipping points, without retraining.
Equal contribution with Charlotte Ricarda Doll.
My master’s thesis explores how Fourier Neural Operators can accelerate simulations of turbulent atmospheric flows. Trained on Large Eddy Simulation data using curriculum learning and generalised teacher forcing, the models reproduce key features of atmospheric turbulence. Training on full three-dimensional data improves accuracy, with an estimated speedup of up to 65× in the evaluated setup. Adding local differential kernels further improves accuracy, but reduces the speedup.
The thesis is available upon request.
Request accessDeveloped in collaboration with the Australian National University (ANU) in Canberra, my bachelor’s thesis explores how machine learning can reduce the time spent aligning lasers. It combines image-based mode classification, neural-network optimisation and computer-controlled actuators to align laser beams to optical resonators. The toolkit achieved mode-matching efficiencies of up to 88.4% of the physically achievable limit and reduced the need for human intervention, although resonator locking still required manual calibration.
The thesis is available upon request.
Request accessBeyond research
I’m passionate about scouting, and I play and train unicycle hockey through Heidelberg University’s Hochschulsport.
Apart from that, I love spending time outdoors, reading a good book, or building a small website for fun (see projects below).
A travel journal from my year in Australia, 2021–2022.
A local guide to restaurants, bakeries and hairdressers in Dossenheim.
Discover the bars, pubs and cafés on Heidelberg’s Untere Straße.
A collection of games I built during the COVID lockdowns.
An organisational tool for sharehouses to distribute household chores, manage shared expenses and keep a shared shopping list.
The scouting website I built.
Get in touch
Questions about my research, shared interests or just saying hello — I’d be happy to hear from you.
gt04(at)duck.com