Niclas focuses on developing principled approaches to make large-scale machine learning systems more robust, interpretable, and trustworthy. His research spans both theoretical foundations and new method development. On the theoretical side, he has worked with Geoff Pleiss at the Vector Institute to investigate ensemble properties in the overparameterized regime and explored universal approximation theory for efficient transformers. On the methodological side, he has developed new approaches for scalable Boltzmann sampling using energy-weighted flow matching and collaborated with Niki Kilbertus at the Helmholtz AI Institute on estimating nonlinear causal effects.