Title:Markov adaptive Pólya trees and multi-resolution adaptive shrinkage in nonparametric modeling

Abstract:We introduce a hierarchical nonparametric model for probability measures based on a multi-resolution transformation of probability distributions. The model allows a varying amount of shrinkage to be applied to data features of different scales and/or at different locations in the sample space, and the varying shrinkage level is locally adaptive to the empirical behavior of the data. Moreover, the model's hierarchical design---through a latent Markov tree structure---allows borrowing of information across locations and scales in setting the adaptive shrinkage level. Inference under the model proceeds efficiently using general recipes for conjugate hierarchical models. We illustrate the work of the model in density estimation and evaluate its performance through simulation under several schematic scenarios carefully designed to be representative of a variety of applications. We compare its performance to those of several state-of-the-art nonparametric models---the Pólya tree, the optional Pólya tree, and the Dirichlet process mixture of normals. In addition, we establish several important theoretical properties for the model including absolute continuity, full nonparametricity, and posterior consistency.