“…Such compact representation which retains most important information of a high-dimensional matrix can provide a significant reduction in memory requirements, and more importantly, computational costs when the latter scales, e.g., according to a high-degree polynomial, with the dimensionality. Matrices with low-rank structures have found many applications in background subtraction [1], [2], [3], [4], system identification [5], IP network anomaly detection [6], [7], latent variable graphical modeling, [8], ranking and collaborative filtering, [9], subspace clustering [10], [11], [12], adaptive, sensor and multichannel signal processing [13], [14], [15], [16], [17], [18], [19], [20], [21], [22], [23], [24], [25], [26], [27], [28], [29], [30], [31], biometrics [32], [33], statistical process control and multidimensional fault identification [34], [35], quantum state tomography [36], and DNA microarray data [37]. Singular value decomposition (SVD) [38] and the rankrevealing QR (RRQR) decomposition [39], [40] are among the most commonly used algorithms for computing a lowrank approximation of a matrix.…”