A Collective Neurodynamic Optimization Approach to Nonnegative Tensor Decomposition

In this paper, a collective neurodynamic optimization approach is proposed to nonnegative tensor factorization. Tensor decompositions are often applied in the data analysis. However, it is often a nonconvex optimization problem, which would cost much time and usually trap into the local minima To solve this problem, a novel collective neurodynamic optimization approach is proposed by combining recurrent neural networks (RNN) and particle swarm optimization (PSO) algorithm. Each RNN still carries out local search. And then the best solution of each RNN improves through PSO framework. In the end, the global optimal solutions of nonnegative tensor factorization are obtained. Experiments results demonstrate the effectiveness for the nonconvex optimization with constraints.