B1281
Title: Recursive estimation of probability distributions
Authors: Lorenzo Cappello - Universitat Pompeu Fabra (Spain) [presenting]
Stephen Walker - University of Texas at Austin (United States)
Abstract: The purpose is to discuss a family of recursive algorithms defining a sequence of probability distributions. The motivating application is offered by existing iterative schemes related to Bayesian predictive updates, particularly the predictive distributions of Dirichlet Process mixtures. The weak convergence of the sequence is established, stating the problem as a fixed-point estimation of an infinite-dimensional function and sufficient conditions for convergence are discussed. Convergence of existing and new algorithms is established using the presented result. Finally, empirical application performance, such as regression and inverse problems, is illustrated.