Postdoctoral researcher · University of Tübingen
I develop probabilistic methods for simulation and inference in dynamical systems, from Kalman filters in millions of dimensions to generative models of the climate. Lately, I am drawn to a new question: what these tools can tell us about social dynamics.
Together with Philipp Hennig and the Methods of Machine Learning group in Tübingen.
Bayesian filtering and smoothing for high-dimensional systems, and probabilistic numerical solvers for ODEs and PDEs. Exploiting structure (low rank, Kronecker, sparsity) allows scaling principled uncertainty quantification to real-world problems.
Diffusion models for simulation generates output that comes with uncertainty instead of yielding a point estimate. Practical example: turn coarse climate simulations into spatiotemporally coherent, probabilistic weather-scale fields.
A new direction, currently in the process of sharpening the focus. Currently investigating heterogeneous-agent models of wealth and income, and the (stochastic-process) machinery behind them. I am interested both in the application and in the methodology; and happy to chat about both, anytime!
