Online Program Home
My Program

Abstract Details

Activity Number: 573 - Simulation and Stochastic Bayesian Modeling
Type: Contributed
Date/Time: Wednesday, July 31, 2019 : 2:00 PM to 3:50 PM
Sponsor: Section on Bayesian Statistical Science
Abstract #305208
Title: Dirichlet Process Gaussian Process Model for Photometric Redshift
Author(s): Arindam Fadikar* and David Higdon and Jonas Chaves-Montero and Salman Habib
Companies: and Virginia Tech and Argonne National Lab and Argonne National Lab
Keywords: bayesian inference; gaussian process; dirichlet process; photometric redshift; mcmc

Gaussian process (GP) model has been proven effective in modeling photometric redshift as function of flux. However, such model suffers from a number of issues: poor predictibility at extreme redshifts, inability to account for arbitrary uncertainty and proper management of outliers. Instead, a GP mixture model based on spatial dirichlet process solves a number of issues. Motivating by this example, we develop a novel non-parametric spatial modeling technique based on Dirichlet Process - Gaussian Process (DPGP), which models data distribution as mixture of normals, allowing the mixting proportion to change over input space. This modeling approach uses more than one GP to model the data by discovering latent cluster distribution using a Dirichlet Mixture model. We show illustrations of DPGP model using a simulated and a real example from redshift estimation problem.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2019 program