A recurrent neural network for real-time matrix inversion (Q2365631)

A linear recurrent neural network is proposed for large-scale nonsingular matrix inversion in real-time. The proposed recurrent neural network for \(n \times n\) matrix inversion consists of \(n^ 2\) neurons (processing elements) arranged spatially in a matrix form and can be decomposed into \(n\) independent subnetworks. The asymptotic stability of the activation states and the capability of the recurrent neural network to generate inverse matrices within a few time constants are proved.