Relative degree of recurrent neural networks

In the paper some tools from the differential geometry theory of single-input single-output nonlinear systems are applied to a recurrent neural network. It is shown that a change of coordinates and a state feedback can transform a recurrent neural network in to a linear system.<
>