Speaker: Chris Wikle, University of Missouri
Abstract: Spatio-temporal data are ubiquitous in the sciences, medicine, and engineering, and their study is important for understanding and predicting a wide variety of processes. One of the difficulties with statistical modeling of spatial processes that change in time is the complexity of the dependence structures that must describe how such a process varies, and the presence of high-dimensional complex datasets and large prediction domains. It is particularly challenging to specify parameterizations for nonlinear dynamic spatio-temporal models (DSTMs) that are simultaneously useful scientifically, efficient computationally, and allow for proper uncertainty quantification. Here I describe recent approaches that fuse physically plausible mechanistic relationships as well as neural reservoir components to facilitate modeling the evolution of complex nonlinear spatio-temporal processes. In particular, I consider mechanistic level-set dynamics (i.e., modeling the evolution of bounded objects) and utilize echo-state networks to assist with capturing speed of evolution. The motivating problem is concerned with modeling the evolution of large wildfires in complex environments, but this approach can be used to model the change of other objects over time, such as tumor growth or shrinkage under treatment.
Chris will be on-site for the seminar in W2008, but the seminar will also be streamed through Zoom.
ZOOM
Meeting ID: 928 8613 3715
Passcode: 838290
