Greetings! I’m Manuel. I am currently pursuing my Ph.D. in Machine Learning (ML) at the University of Tuebingen under the mentorship of Professor Jakob Macke. My primary academic interest lies in the development of flexible and efficient methods for Bayesian inference, which I apply to various scientific domains.
Currently, my research is centered around the fascinating field of simulation-based inference (SBI). Simulators have become indispensable tools in the realm of scientific exploration. They allow us to replicate natural processes and measurements, providing valuable insights into complex phenomena. Leveraging computer programming and high-performance computing, we can create detailed synthetic data which is instrumental in predicting and understanding scientific and engineering phenomena.
However, inferring unobservable quantities within these simulations presents a significant statistical challenge. Methods from deep learning and generative modeling have become popular for addressing such inverse problems. Nonetheless, they can be sensitive to model misspecification or even minor adversarial perturbations. Furthermore, they often necessitate a substantial amount of training data, which is inherently limited for (slow!) scientific simulators. These are some of the problems I am actively working to address through my research.