On the Product Probability Kernel in Sparse Gaussian Process Models

Bjørn Sand Jensen, Jens Brehm Nielsen, Jan Larsen

AbstractIn this report we consider the particular use of the product probability kernel in models based on the pseudo-input formulation of Gaussian process priors. In particular we focus on modeling inputs as generative mixture models which in effect “forces” the pseudo-inputs to be densities as well. We focus on the learning and interpretation of these pseudo-input/densities in a clustering sense, and provide a few connections with other formulations of covariance structure for pseudo inputs.
KeywordsProduct probability kernel, Kernel Methods, Gaussian Processes, Clustering, Generative Kernels
TypeMisc [Other]
Year2011    Month October
PublisherDTU - IMM
BibTeX data [bibtex]
IMM Group(s)Intelligent Signal Processing


Back  ::  IMM Publications