Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP)This paper presents an algorithm for signal subspace separation in the context of multidimensional data. The proposal is an extension of the randomized Singular Value Decomposition (SVD) for higher-order tensors. From a set derived from random sampling, we construct an orthogonal basis associated with the range of each mode-space of the input data tensor. Multilinear projection of the input data onto each mode-space then transforms the data to a low-dimensional representation. Finally, we compute the Higher-Order Singular Value Decomposition (HOSVD) of the reduced tensor. Furthermore, we propose an algorithm for computing the randomized HOSVD based on the row-extraction technique. The results reveal a relevant improvement from the standpoint of computational complexity.215219Fundacao de Amparo a Pesquisa do Estado de Sao Paulo (FAPESP), Brazil [2014/23936-4]Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP)24th European Signal Processing Conference (EUSIPCO)AUG 28-SEP 02, 2016Budapest, HUNGARY