EcoSta 2018: Registration
View Submission - EcoSta2018
Title: Variance change point detection for data on a surface Authors:  Zhenguo Gao - Virginia Tech (United States)
Pang Du - Virginia Tech (United States) [presenting]
Abstract: Motivated from an organ procurement application, we consider the problem of variance change point for data on a surface. This change point would suggest a deterioration of the organ to the non-viable status. The statistical challenge here is the development of an efficient procedure that can simultaneously estimate a smooth mean trend on a 2D surface and detect the change point in the variance function on the surface. A naive adoption of the existing methods can result in substantial computational difficulty since the data were collected at a dense grid on a 2D surface. We devise an efficient method that combines subsampling with thin-plate spline smoothing and variance change point detection. Simulations are performed to verify its empirical performance and an application to the organ procurement data is provided.