CMStatistics 2018: Start Registration
View Submission - CMStatistics
Title: Sparse and robust composite likelihood inference with application to parcel-based evoked brain activity analysis Authors:  Zhendong Huang - The University of Melbourne (Australia) [presenting]
Davide Ferrari - University of Modena (Italy)
Abstract: Analysing Blood-oxygen-level dependent (BOLD) signal in multiple active regions of brain is a popular and challenging problem in biological study. In an experiment, the brain region of interest is divided into voxels with BOLD signal observed in each voxel through time. Classical methods for estimating time series suffer from loss of efficiency due to the high-dimension of the problem and the absence of knowledge on the correlation structure between voxels. An improved composite likelihood method will be introduced to give inference on BOLD signal. The new method seeks sparse composition rule to include only a small proportion of voxels, while obtaining efficiency to the largest extent in the final estimation. Performance of the new method will be illustrated through theoretical results, numerical studies and an application to real BOLD signal data.