Spectral Mixture Kernel Approximation Using Reparameterized Random Fourier Feature

Yohan Jung, Jinkyoo Park

Keywords: Gaussian Process, Spectral Mixture Kernel, Random Fourier Feature, Sparse Spectrum Approximation, Stochastic Gradient Variational Bayes, Natural Parameter

Abstract: We propose a method for Spectral Mixture kernel approximation using the Reparameterized Random Fourier Feature (R-RFF) in the sense of both general parameter and natural parameter view. Meanwhile, we provide the effective sampling methods of spectral points which samples the number of spectral points by considering the normalized weight parameters of SM kernel. Also, we develop the regularized sparse spectrum approximation by using Stochastic Gradient Variational Bayes for scalable learning of GP model with SM kernel.