EigenGP: Gaussian processes with sparse data-dependent eigenfunctions

Details EigenGP: Gaussian processes with sparse data-dependent eigenfunctions EigenGP: Gaussian processes with sparse data-dependent eigenfunctions

2012-04-18

arxiv.org/abs/1204.3972v1

Computer Science

Learning

13 pages, 10 figures

Scientific paper

Gaussian processes (GPs) provide a nonparametric representation of functions. Given N training points, the exact GP inference incurs high computational cost. A variety of sparse GP methods have been proposed to speed up GP inference. These methods essentially trade prediction accuracy with computational effciency. In this paper, we address this problem from a new perspective: we define an exact Gaussian process model, EigenGP, whose covariance function has a sparse spectrum adaptive to data. Specifically, we estimate eigenfunctions of covariance function based on training data and use an empirical Bayesian approach to select these eigenfunctions. Thus, unlike the previous Nystrom-based methods, EigenGP defines an exact Gaussian process model with an data-dependent covariance function. To handle nonlinear likelihoods, we develop an efficient expectation propagation inference algorithm, and couple it with an active-set algorithm for evidence maximization. Because the selected eigenfunctions (based on Gaussian kernels) are naturally associated with data clusters, EigenGP is also suitable for semi-supervised learning. Our experimental results demonstrate improved predictive performance of EigenGP over alternative methods for classification tasks.

Affiliated with

Also associated with

No associations

Rating

EigenGP: Gaussian processes with sparse data-dependent eigenfunctions does not yet have a rating. At this time, there are no reviews or comments for this scientific paper.

If you have personal experience with EigenGP: Gaussian processes with sparse data-dependent eigenfunctions, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and EigenGP: Gaussian processes with sparse data-dependent eigenfunctions will most certainly appreciate the feedback.

Rate now

Comments { 0 }

Profile ID: LFWR-SCP-O-410999