Separable nonlinear least-squares methods for on-line estimation of neural nets Hammerstein models

Isometric muscle contraction induced by repetitive peripheral magnetic stimulation (RPMS)—Modeling and identification

“…(22) at every iteration by means of a recursive least squares algorithm. For an extensive derivation the reader is referred to [21,19].…”

Highly efficient parameter estimation algorithms for Hammerstein non‐linear systems

“…In addition to the contributions mentioned above, a lot of other works on Hammerstein system identification exist in the literature. For example, Ngia and Westwick et al applied the separable non‐linear least squares optimisation methods in [35] to identify the Hammerstein models [36, 37], Goethals et al [38] presented a subspace identification method for Hammerstein systems using the least squares support vector machines.…”

Variable projections neural network training

“…Besides the obvious advantage of reducing the number of parameters to be determined by eliminating the linear parameters, it has been shown [30,33,22,24] that the resulting reduced problem is better conditioned than the original full one, and if the same optimization algorithm is used it always converges in less iterations. Since with a careful implementation, the cost per iteration is about the same, these results and extensive applications as reported in [14] (where a wealth of references can be found) indicate that there is a net gain in using VP over solving the unreduced problem with a conventional non-linear least squares algorithm.…”