Nonlinear Functions Activated Gradient-based Neural Dynamics for Online Matrix Inversion

In an effort to improve the convergence performance of gradient-based neural dynamics (GND), by drawing upon two novel activation functions (AFs), a fixed-time convergent nonlinear GND (NLGND) model is proposed for online solving the problem of matrix inversion. In comparison to the existing GND models under the action of other AFs (e.g., the linear function, the power-sigmoid function, and the sign-bi-power function), the superiority of the proposed NLGND model lies in its rapid fixed-time convergence, which is simulatively verified by two illustrative examples.