Introduction to Gaussian Processes, Modeling Icy Terrain with Gaussian Processes, and a Framework for Exploiting Periodic Data

Kyle Schmitt
Seminar

This seminar will serve as an introduction to Gaussian processes methodology and its advantages in spatial uncertainty quantification. The discussion will aim to present Gaussian processes both theoretically and technically. The entire progression of the method will be presented including the evaluation of the hyper-parameters from data, the computation of the posterior distribution characteristics at new sample point from the data, and the multivariate Gaussian sampling. After the method is presented, an application will be presented demonstrating the modeling of an icy road with Gaussian processes (and perhaps the simulation of a vehicle on the icy road). If time permits, a Gaussian processes framework for periodically distributed data will be presented; the framework leverages the covariance matrices' periodic structure to allow for Fast Fourier transform evaluation resulting in greatly reduced simulation times.