Recurrent neural networks for solving linear matrix equations (Q1311981)

The author describes how recurrent neural networks can be used for solving linear matrix equations. He shows that the methods are asymptotically stable in the large and can be used for solving Lyapunov equations. He gives two examples for handling continuous-time and discrete-time Lyapuov equations. The author also shows the relationship between the Lyapunov function and the activation function. The proposed recurrent neural networks are better than the supervised learning approach, since prior training is unnecessary.