Online Program Home
  My Program

Abstract Details

Activity Number: 154 - Statistical and Imaging Methods in Astronomy and Astrophysics
Type: Topic Contributed
Date/Time: Monday, July 31, 2017 : 10:30 AM to 12:20 PM
Sponsor: Astrostatistics Special Interest Group
Abstract #323820
Title: Photometric Classification of Supernova with Biased Training Set
Author(s): David Van Dyk* and Roberto Trotta and Esben Revsbech
Companies: Imperial College London and Imperial College London and Imperial College London
Keywords: Bayesian Methods ; Classification ; Propensity Scores ; Two-Stage Analysis

The expansion of the Universe can be studied via a Bayesian model that relates the difference between the apparent and intrinsic brightnesses of objects to their distance which in turn depends on parameters that describe this expansion. While apparent brightness can be readily measured, intrinsic brightness can only be obtained for certain objects. Type Ia Supernova (SNIa) occur when material accreting onto a white dwarf triggers a powerful supernova explosion. Because this occurs only in a particular physical scenario, we can estimate the intrinsic brightness of SNIa. To take advantage of this, however, SNIa must be precisely classified using (low resolution) photometric data and a biased training set. We use Gaussian Processes to account for irregular observation times and diffusion maps to identify features for a random forest classifier. To account for bias in the overall training set, we use propensity scores to form homogeneous groups where the training subsets are more representative. Finally we enrich the training sets by probabilistically generating synthetic data. In this way we are able to identify SNIa nearly as well as we would with an unbiased training set.

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

Back to the full JSM 2017 program

Copyright © American Statistical Association