Total: 1
The canonical polyadic tensor decomposition has a long history. But it becomes challenging to find a tensor decomposition when the rank is between the largest and the second-largest dimension. In such cases, traditional optimization methods, such as nonlinear least squares or alternative least squares methods, often fail to find a tensor decomposition. There are also direct methods, such as the normal form algorithm and the method by Domanov and De Lathauwer, that solve tensor decompositions algebraically. However, these methods can be computationally expensive and require significant memory, especially when the rank is high. This paper proposes a novel two-stage algorithm for the order-3 nonsymmetric tensor decomposition problem when the rank is not greater than the largest dimension. It transforms the tensor decomposition problem into two optimization problems. When the first-stage optimization is not fully solved, the partial solution will also be leveraged in the second-stage optimization problem. We prove the equivalence between tensor decompositions and the global minimizers of the two-stage optimization problems. Our numerical experiments demonstrate the proposed two-stage optimization method is very efficient and robust, capable of finding tensor decompositions where other commonly used state-of-the-art methods fail.