Standing Wave Decomposition Gaussian Process

  • 2018-05-24 15:56:06
  • Chi-Ken Lu, Scott Cheng-Hsin Yang, Patrick Shafto
  • 0

Abstract

We propose a Standing Wave Decomposition (SWD) approximation to GaussianProcess regression (GP). GP involves a costly matrix inversion operation, whichlimits applicability to large data analysis. For an input space that can beapproximated by a grid and when correlations among data are short-ranged, thekernel matrix inversion can be replaced by analytic diagonalization using theSWD. We show that this approach applies to uni- and multi-dimensional inputdata, extends to include longer-range correlations, and the grid can be in alatent space and used as inducing points. Through simulations, we show that ourapproximate method outperforms existing methods in predictive accuracy per unittime in the regime where data are plentiful. Our SWD-GP is recommended forregression analyses where there is a relatively large amount of data and/orthere are constraints on computation time.

 

Quick Read (beta)

loading the full paper ...