The Role of Bayesian Analysis for an Emerging Class of Complex Data: Object Data

• Activity Number: 383
• Type: Roundtables
• Date/Time: Tuesday, August 6, 2013 : 12:30 PM to 1:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Jeffrey S. Morris*+
• Companies: The University of Texas MD Anderson Cancer Center
• Keywords: Bayesian methods ; Big data ; Functional Data Analysis ; Image Data ; Hierarchical Models ; Object Data

Abstract

An increasing number of scientific studies yield complex, structured, high-dimensional data, much of which could be characterized as "object data," to use the term coined by Steve Marron for the 2010--2011 SAMSI program of that name. Object data involve multiple measurements on some structured space, including but not limited to functions, images, shapes, graphs, and trees. The internal structure of these objects can be based on geometry or application-specific scientific relationships. One of the key challenges in these data is to combine information across objects to draw population inference while also accounting for the internal structure within the objects. There are often considerable computational hurdles to overcome, given the complexity and high dimensionality typical of these data. The Bayesian paradigm is uniquely suited for this challenge, with hierarchical models, flexible modeling tools, sparsity priors to induce parsimony, and a vast computational toolbox. In this roundtable, we will talk about this emerging area of statistics and discuss how Bayesian researchers can contribute to the effort of developing computational tools for this broad class of data.