Higher order tensor model has been seen as a potential mathematical framework to manipulate the multiple factors underlying the observations. In this paper, we propose a flexible two stage algorithm for K-mode Convolutive Nonnegative Tucker Decomposition (K-CNTD) model by an alternating least square procedure. This model can be seen as a convolutive extension of Nonnegative Tucker Decomposition (NTD). Shift-invariant features in different subspaces can be extracted by the K-CNTD algorithm. We impose additional sparseness constraint on the algorithm to find the part-based representations. Extensive simulation results indicate that the K-CNTD algorithm is efficient and provides good performance for a feature extraction task.
展开▼
