DMCA
Bayesian Gaussian Process Models: PAC-Bayesian Generalisation Error Bounds and Sparse Approximations (2005)
| Citations: | 72 - 14 self |
BibTeX
@TECHREPORT{Seeger05bayesiangaussian,
author = {Matthias Seeger},
title = {Bayesian Gaussian Process Models: PAC-Bayesian Generalisation Error Bounds and Sparse Approximations},
institution = {},
year = {2005}
}
Share
 |
 |
 |
 |
||
OpenURL
Abstract
Non-parametric models and techniques enjoy a growing popularity in the field of machine learning, and among these Bayesian inference for Gaussian process (GP) models has recently received significant attention. We feel that GP priors should be part of the standard toolbox for constructing models relevant to machine learning in the same way as parametric linear models are, and the results in this thesis help to remove some obstacles on the way towards this goal. In the first main chapter, we provide a distribution-free finite sample bound on the difference between generalisation and empirical (training) error for GP classification methods. While the general theorem (the PAC-Bayesian bound) is not new, we give a much simplified and somewhat generalised derivation and point out the underlying core technique (convex duality) explicitly. Furthermore, the application to GP models is novel (to our knowledge). A central feature of this bound is that its quality depends crucially on task knowledge being encoded
Keyphrases
sparse approximation pac-bayesian generalisation error bound bayesian gaussian process model machine learning significant attention gaussian process generalised derivation pac-bayesian bound gp model convex duality central feature gp prior parametric linear model first main chapter task knowledge gp classification method distribution-free finite sample bound bayesian inference ii non-parametric model standard toolbox general theorem core technique thesis help
