Gaussian process classification: a message-passing viewpoint
AUTHOR:
Filipe Rodrigues (fmpr [at] dei.uc.pt)
ABSTRACT:
The goal of this tech report is to provide a message-passing viewpoint of the Expectation Propagation (EP) algorithm commonly used for Gaussian process (GP) classification with probit likelihood. By presenting this EP algorithm as message-passing in the factor graph gives the reader a different (more unified) perspective on what the algorithm is doing and facilitates the use of GP classification (and regression) as a building block for larger factor graphs.