A Support Set Selection Algorithm for Sparse Gaussian Process Regression

Abstract

Gaussian process is difficult to apply to the large data due to its computational problem. Many sparse methods have been proposed to deal with this problem. The majority focus on regression by a small size of support set. In this paper, we aim to propose a simple and efficient support set selection algorithm for Gaussian process regression. We describe a new selection criterion based on residual sum of squares to score the importance of training data and then update the support set iteratively according to this score. However, the iterative updating procedure has high time complexity due to the re-computing of matrix. Then we further speed up the selection algorithm based on some matrix operation.