State-space approach to continuous recurrent neural networks
R. Żbikowski · 2003
Continuous-time recurrent neural schemes are presented in the context of the state-space approach to nonlinear identification and control. Recent learning algorithms are evaluated from the control and identification viewpoint. The issues of stability, convergence and persistent excitation are addressed, and a precise definition of the generalization property is given. The notion of neural nonlinear adaptive control is introduced.>