Statistics Seminars: Local Bayesian Principal Curves
22 January 2008 14:15 in CM221
This talk is concerned with Local Bayesian Principal Curves, a Bayesian, piecewise linear model for principal curves. The talk first of all reviews various definitions of principal curves, presents the corresponding algorithms and discusses their shortcomings. Most of these shortcomings can be overcome by the Local Bayesian Principal Curve model. An MCMC algorithm for sampling from the posterior distribution of the curves is presented. Generalisations to the case of intersecting and unconnected curves are presented as well as the generalisation of the principal curve algorithm to principal manifolds.
Contact email@example.com for more information