PREDICTION WITH GAUSSIAN PROCESSES: BASIC IDEAS AND NEW DIRECTIONS
Dr Chris Williams, Institute for Adaptive and Neural Computation, Division of Informatics, University of Edinburgh

Abstract: Gaussian Process predictors are a kernel machine prediction method based on a Gaussian process (GP) prior over functions. In the first part of the talk I will describe the basic ideas of GP prediction for regression and classification problems. I will then go on to discuss some more recent topics such as approximation methods for large datasets, the work of M. Seeger (Edinburgh) on PAC-Bayesian bounds for GP classifiers, and information-theoretic characterization of learning curves for Gaussian Processes.

This seminar was held at the Department of Computer Science, Royal Holloway, University of London on 15 January 2002.