Adaptive vs. accommodative neural networks for adaptive system identification

Adaptive multilayer perceptrons (MLPs) with long- and short-term memories (LASTMs) and accommodative MLPs with interconnected neurons (MLPWINs) have been mathematically justified for adaptive processing. The benefits of using these neural networks for adaptive processing include less online computation, no poor local extrema to fall into, and much more timely and better adaptation as compared with using neural networks with all their weights adjusted online for adaptation. In this paper, adaptive MLPs with LASTMs and accommodative MLPWINs are compared for adaptive identification of dynamical systems in the series-parallel formulation. Numerical examples show that adaptive MLPs with LASTMs have much better generalization ability than accommodative MLPWINs.