Moving Least Squares Regression for High-Dimensional Simulation Metamodeling

Moving least squares regression works well as a smoothing method in low-dimensional problems when there are many design points. We will present an algorithm for moving least squares that handles the challenges that arise in high-dimensional problems. We will show applications of the method to stochastic simulation metamodeling, in which simulations are noisy but cheap. In these applications, there can be very many design points, and smoothing mitigates the effect of noise.