Exploring Robust Uncertainty Bounds in RKHS through Convex Optimization

Abstract:

In this seminar, we discuss the findings of the paper "Robust Uncertainty Bounds in Reproducing Kernel Hilbert Spaces: A Convex Optimization Approach".
Specifically, we explore the challenge of establishing out-of-sample bounds for unknown functions in RKHS.
By employing convex optimization techniques, we can compute robust uncertainty bounds even when data is corrupted by bounded noise from any compact distribution.
We will discuss the properties and assumptions of our method, compare it with other approaches, and present numerical experiments to illustrate its effectiveness in various scenarios. Furthermore, we will consider the potential of our method for robustness in predictive modeling without relying on independent data assumptions.”
 

Biography:
Dr. Paul Scharnhorst is currently an R&D Engineer at CSEM SA, Switzerland. He received his Ph.D. from EPFL in collaboration with CSEM in 2024, where he focused on uncertainty quantification, benchmarking, and control for energy systems. Prior to that, he obtained the B.Sc. and M.Sc. degree in Mathematics from the Otto-von-Guericke University Magdeburg, Germany. In 2024, he received the IEEE CSS Swiss Chapter Young Author Best Journal Paper Award for the paper "Robust Uncertainty Bounds in Reproducing Kernel Hilbert Spaces: A Convex Optimization Approach".”