Nonlinearly activated neural network for solving dynamic complex-valued matrix pseudoinverse

A special class of recurrent neural network, termed Zhang neural network (ZNN), has been recently proposed for solving various dynamic problems, and has shown excellent performance in the real-valued domain. In this paper, a new complex-valued ZNN model (termed CVZNN model) is firstly proposed and investigated for online solution of dynamic complex-valued matrix pseudoinverse. Particularly, a novel activation function, called Li activation function, is employed, which is proven to enable the CVZNN model to converge in finite time. For comparative purposes, the linear activation function is exploited for solving such a dynamic complex-valued problem. Computer simulations are conducted to evaluate and compare the performance of CVZNN model with different activation functions for the dynamic complex-valued matrix pseudoinversion. Both theoretical analysis and simulation results verify the efficacy of the CVZNN model with nonlinear activation function.