Online Updating for Gaussian Process Learning
Hongjun Su, Hong Zhang · 2017
Gaussian processes for regression and classification have become an effective machine learning methodology with a number of distinctive advantages. One notable disadvantage of Gaussian process methods is the computational complexity related to the inversion of matrices, especially for applications that involve large datasets. In this paper, an exact online updating algorithm is presented to significantly reduce the amount of computations for repeated progressive trainings of Gaussian processes.