1: \begin{abstract}

2:   We develop a simulation-based method for the online updating of

3:   Gaussian process regression and classification models.  Our method

4:   exploits sequential Monte Carlo to produce a fast sequential design

5:   algorithm for these models relative to the established MCMC

6:   alternative.  The latter is less ideal for sequential design since

7:   it must be restarted and iterated to convergence with the inclusion

8:   of each new design point.  We illustrate some attractive ensemble

9:   aspects of our SMC approach, and show how active learning heuristics

10:   may be implemented via particles to optimize a noisy function or to

11:   explore classification boundaries online.

12: 

13:   \bigskip

14:   \noindent {\bf Key words:} Sequential Monte Carlo,

15:   Gaussian process, nonparametric regression and classification,

16:   optimization, expected improvement, sequential design, entropy

17: \end{abstract}