On Adaptive PARAFAC Decomposition of Three-Way Tensors
Abstract
We propose a fast adaptive PARAFAC algorithm for a class of third-order tensors which have one dimension growing linearly with time. The proposed method is based on alternating least squares approach used in conjunction with a Newton-type optimization technique. By exploiting the Khatri-Rao product and the reduced rank structure at each time instant, our algorithm has the advantages of both linear complexity and superior convergence performance. A modified version of the algorithm is also proposed to tackle the nonnegative adaptive PARAFAC problem. Parallel implementation issues and algorithm performance are finally investigated. The latter is achieved through simulation comparison with the state-of-the-art algorithms to better highlight the effectiveness of the proposed method.
Origin : Files produced by the author(s)
Loading...