Quickest Detection for Changes in Maximal kNN Coherence of Random Matrices (1508.04720v3)

Abstract: This paper addresses the problem of quickest detection of a change in the maximal coherence between columns of a $n\times p$ random matrix based on a sequence of matrix observations having a single unknown change point. The random matrix is assumed to have identically distributed rows and the maximal coherence is defined as the largest of the $p \choose 2$ correlation coefficients associated with any row. Likewise, the $k$ nearest neighbor (kNN) coherence is defined as the $k$-th largest of these correlation coefficients. The forms of the pre- and post-change distributions of the observed matrices are assumed to belong to the family of elliptically contoured densities with sparse dispersion matrices but are otherwise unknown. A non-parametric stopping rule is proposed that is based on the maximal k-nearest neighbor sample coherence between columns of each observed random matrix. This is a summary statistic that is related to a test of the existence of a hub vertex in a sample correlation graph having a degree at least $k$. Performance bounds on the delay and false alarm performance of the proposed stopping rule are obtained in the purely high dimensional regime where $p\rightarrow \infty$ and $n$ is fixed. When the pre-change dispersion matrix is diagonal it is shown that, among all functions of the proposed summary statistic, the proposed stopping rule is asymptotically optimal under a minimax quickest change detection (QCD) model as the stopping threshold approaches infinity. The theory developed also applies to sequential hypothesis testing and fixed sample size tests.