IGPP is pleased to invite you to join its Virtual Seminar Series presentation featuring Matthias Morzfeld, IGPP's newest faculty memeber. Dr. Morzfeld's talk, "Numerical challenges in Bayesian inference: Ill conditioned problems, high-dimensional problems, and trans-dimensional problems," will be available via Zoom on Tuesday, April 28, starting at 12:30pm. Zoom registration is available through here or through the event flyer here.
Tuesday, April 28, 2020
Time: 12:30 pm - 1:45 pm, Pacific Time
Note: This meeting will be recorded.
Abstract: I will provide a quick review of the problem formulation of Bayesian inference and associated
sampling methods. I then discuss the difficulties one can run into when solving Bayesian inference problems via sampling. These difficulties include anisotropies in posterior distributions and, of course, the extremely high-dimensionality typical of many geophysical problems. I will outline theory and intuition about how to overcome these difficulties. Finally, I discuss transdimensional inference in which the complexity of the numerical model is itself an unknown and present preliminary result of an ensemble sampler for trans-dimensional sampling problems.