A new least squares-based approach for fast learning in recurrent neural networks

In this paper a new approach to learning in recurrent neural networks is presented. The method proposed is based on the descent of the error functional in the space of the linear part of the neurons (neuron space approach). A linaer system is solved for the weights following a Recursive Least Squares criterion at each step of the learning process. This approach, w.r.t. traditional gradient-based algorithm, guarantees better performances from the point of view of the speed of convergence and the numerical robustness.