Publications
8. Geometry and convergence of natural policy gradient methods
Joint work with Guido Montúfar, Information Geometry, 2024
7. Achieving High Accuracy with PINNs via Energy Natural Gradients
Joint work with Marius Zeinhofer, International Conference on Machine Learning, 2023
6. Algebraic Optimization of Sequential Decision Problems, preprint openly available
Joint work with Mareike Dressler, Marina Garotte-Lopéz, Guido Montúfar, and Kemal Rose, Journal of Symbolic Computation, Volume 121, 2022
5. Error Estimates for the variational training of neural newtorks with boundary penalty
Joint work with Marius Zeinhofer, Third Conference on Mathematical and Scientific Machine Learning, 2022
4. Notes on Exact Boundary Values in Residual Minimisation
Joint work with Marius Zeinhofer, Third Conference on Mathematical and Scientific Machine Learning, 2022
3. Uniform Convergence Guarantees for the Deep Ritz Method for Nonlinear Problems
Joint work with Patrick Dondl and Marius Zeinhofer, Advances in Continuous and Discrete Models, 2022
2. Invariance properties of the natural gradient in overparametrised systems
Joint work with Jesse van Oostrum and Nihat Ay, Information Geometry, 2022
1. The Geometry of Memoryless Stochastic Policy Optimization in Infinite-Horizon POMDPs
Joint work with Guido Montúfar, Tenth International Conference on Learning Representations, 2022
Preprints
- Optimization in SciML – A Function Space Perspective
Joint work with Marius Zeinhofer
Extended Abstracts
Solving infinite-horizon POMDPs with memoryless stochastic policies in state-action space, with Guido Montúfar, 5th Multi-disciplinary Conference on Reinforcement Learning and Decision Making (RLDM 2022), fulltext available
Deep Ritz revisited, with Marius Zeinhofer, ICLR workshop on Integration of Deep Neural Models and Differential Equations, fulltext available
On the space-time expressivity of residual networks, ICLR workshop on Integration of Deep Neural Models and Differential Equations, fulltext available
Thesis
During my MSc studies at the University of Warwick I was supervised by Nikos Zygouras and Theo Damoulas and completed my thesis on the parameter estimation of determinantal point processes. You can find the PDF here.
At the University of Freiburg, I studied the approximation capabilities of deep residual networks under the supervision of Philipp Harms and the thesis is available here.