ABSTRACT
We review some preliminary results in fitting maps between dynamical systems constructed by embedding experimental time series. We describe two applications of the Radial Basis Function (RBF) network trained with Least Squares error minimisation: in iteratively predicting a time series by modelling its generating dynamics; and, combined with a linear filter, in separating nonlinear noise from a corrupted signal. We then discuss a symmetrical modification of the RBF network trained with Total Least Squares error minimisation, and its application in testing for differentiable equivalence between embedded systems.
