UC Irvine

In this paper, we propose a novel regularized mixture model for clustering matrix-valued image data. The new framework introduces a sparsity structure (e.g., low rank, spatial sparsity) and separable covariance structure motivated by scientific interpretability. We formulate the problem as a fi-nite mixture model of matrix-normal distributions with regularization terms, and then develop an Expectation-Maximization-type of algorithm for efficient computation. Simulation results and analysis on brain signals show the excellent performance of the proposed method in terms of a better prediction accuracy than the competitors and the scientific interpretability of the solution.