Nonparametric Methods for† Curve Fitting and Prediction with Large Data-sets
EPSRC GR/T29253/01 (Jan. 2005óJune 2007)
The overall aim of this project is to develop statistical† methodology and the statistical theory for curve fitting and related problems, focusing particular on the problems with large data-sets and with large dimensional of input covariates, and to apply them to the modelling and control of nonlinear dynamic systems in engineering. We will extend the idea of mixtures and Gaussian processes to cover a much wider class of models and data structures; to explore ways of modelling the mean and covariance structures by combining local parametric and non-parametric approaches; and to develop efficient algorithms suitable for automatic machine learning and further applications. The core statistical techniques used in this project to analyse the batch data will enable new insight into the related engineering problems.