A modal approach for clustering matrices

In this work we propose a modal approach to density-based clustering for matrix-valued data. We introduce appropriate kernels for this type of data and define a kernel estimator of matrix-variate density functions. Additionally, we propose an extension of the mean-shift procedure for the identification of the modes of the estimated density. Given the intrinsic high dimensionality of matrix-variate data and the resulting computational complexity of the algorithm, we discuss a possible solution to handle the problem. We finally present the performance of the proposed method through an application to real world data, also with respect to a plausible competitor.