IGPP is pleased to invite you to join its Fall 2022 Seminar Series presentation featuring University of Copenhagen's Klaus Mosegaarl. Dr. Mosegaarl's talk, "Are Monte Carlo Methods useful for Large-Scale Inverse Problems?" will be available via Zoom on MONDAY, October 24, 2022, starting at 12:00pm. Zoom: https://ucsd.zoom.us/j/97645897807?pwd=Q01OMk9ZTjlaSEhrVSt5bFBheG50dz09Meeting Password: igpptalk
Time: 12:00 pm, Pacific Time
Location: Revelle 4000 (on-site with a zoom link)
Abstract: An inverse problem is about inferring information about unknown structure of a physical system from observational data. It can be formulated probabilistically, and if the problem is highly nonlinear, a conceptually simple way of implementing a solution is to use Monte Carlo methods. However, experience has shown that typical implementations of these methods are computationally very time-consuming. This phenomenon is usually attributed to the so-called 'curse of dimensionality', but we shall see that this explanation is superficial and overlooks key factors of the phenomenon. Furthermore, we will see how asymptotically exact Monte Carlo analysis of even large-scale problems can be carried out if simplified physics is used to guide the sampling.