Learning Curves for Gaussian Processes via Numerical Cubature Integration

This paper is concerned with estimation of learning curves for Gaussian process regression with multidimensional numerical integration. We propose an approach where the recursion equations for the generalization error are approximately solved using numerical cubature integration methods. The advantage of the approach is that the eigenfunction expansion of the covariance function does not need to be known. The accuracy of the proposed method is compared to eigenfunction expansion based approximations to the learning curve.

• Type: Journal
• Year: 2011
• Where: ICANN
• Authors: Simo Särkkä
• Keywords: Eigenfunction Expansion, Generalization Error, ICANN 2011, Neural Networks, Recursion Equations