B0748
Title: Sequential change-point detection for correlation matrices
Authors: Liyan Xie - The Chinese University of Hong Kong - Shenzhen (China) [presenting]
Abstract: The quickest change detection for correlation matrices is studied. The pre-change correlation matrix is assumed to be an identity matrix (i.e., no pairwise correlation) and the post-change correlation matrix is unknown. Detection statistics are proposed based on the sample correlation matrices. Two types of detection statistics are studied: one is the sum type statistics, which is good at detecting dense changes; the other is the max type statistics, which is good at detecting sparse changes in the correlation matrix. Different types of detection procedures are conducted in parallel to improve detection efficiency. Theoretical guarantees are provided for the evaluation criteria, including average run length to false alarm and the detection delay. Synthetic and real data examples are also provided to validate the performance.
