Efficient training of neural nets for nonlinear adaptive filtering using arecursive Levenberg-Marquardt algorithm

The Levenberg-Marquardt algorithm is often superior to other training algor ithms in off-line applications. This motivates the proposal of using a recu rsive version of the algorithm for on-line training of neural nets for nonl inear adaptive filtering. The performance of the suggested algorithm is com pared with other alternative recursive algorithms, such as the recursive ve rsion of the off-line steepest-descent and Gauss-Newton algorithms. The adv antages and disadvantages of the different algorithms are pointed out, The algorithms are tested on some examples, and it is shown that generally, the recursive Levenberg-Marquardt algorithm has better convergence properties than the other algorithms.