Personal tools

Methods in Statistical Change-point Analysis

— filed under:

Robert Steward, Saint Louis University

  • PhD Oral Defense
When Fri, Feb 06, 2015
from 09:00 AM to 11:00 AM
Where Ritter Hall 200
Add event to calendar vCal

Abstract:  The statistical change-point problem is considered in a variety of settings. In particular, a number of statistical models are developed that may be applied to the problem of detecting the existence of a statistical change-point and then estimating its location in one and several dimensions. Of primary interest in the univariate case are statistical change-points in mean, variance, and multiple change-points. For the multivariate case these cases are also investigated, but also considered is the question in which dimensions the change-point occurred.  New methods proposed in this dissertation answer these questions by applying a Bayesian statistical analysis on the wavelet detail coefficients after a Discrete Wavelet Transform is taken of the original time series.  This new approach is shown to offer several advantages from previously known classical methods. Furthermore, through the use of random matrix dimension reduction techniques, applications to high dimensional time series are also presented. Finally, the method of Reversible Jump Markov Chain Monte Carlo is applied to the multivariate change-point problem to detect dimensional statistical change-points. Along with detailed derivations of theoretical results, numerical experiments involving both simulated and real world data will be presented.

« November 2017 »
Upcoming Events
Algebra Seminar
Tue, Nov 21, 2017
Additive Representations of Elements in Rings Feroz Siddique, University of Wisconsin-Barron County
Computer Science Seminar
Tue, Nov 21, 2017
Generalized Persistence Models, II by David Letscher, SLU
Previous events…
Upcoming events…