|
Abstract:
|
Gaussian process (GP) models have proven to be very useful in modeling computer simulation output. This is because many computer codes are essentially noiseless and respond very smoothly to changes in input settings. In some applications, the simulation output is a function of a large p-dimensional input vector x. After a fixed number of simulations are carried out, a GP model can be used to predict the simulation output at untried settings. It turns out, that the run sizes required for high dimensional inputs can be computationally burdensome for fitting the GP. In this talk, we describe strategies we've found useful for fitting such models when the number of simulation runs is large. This work is motivated by an ongoing application in cosmology.
|
