Statistics Seminars: Kriging with mixtures
9 November 2011 16:00 in CM221
Motivated by issues in palaeo-climate reconstruction, this paper introduces several methodological novelties to the generic issue of temporal smoothing of data in a Bayesian setting. These methodologies include: the use of data-marginal posteriors as a starting point for smoothing; a fast algorithm, even for multi-modal signals, when these are approximated by finite Gaussian mixtures; the use of a Normal Inverse Gaussian distribution as a prior for a smoother that may respond to multi-modal data-marginal posteriors that exhibit abrupt changes. Additionally, and exploiting the speed, the method deals naturally with temporal uncertainty in the underlying data.
Contact firstname.lastname@example.org for more information