Information Sciences: an International Journal
| Hi-index | 0.00 |
A mixture of experts (ME) model provides a modularapproach wherein component neural networks are madespecialists on subparts of a problem. In this framework,that follows the "divide-and-conquer" philosophy, agating network learns how to softly partition the inputspace into regions to be each properly modeled by one ormore expert networks. In this paper, we investigate theapplication of different ME variants to some multivariatenonlinear dynamic systems identification problems whichare known to be difficult to be dealt with. The aim is toprovide a comparative performance analysis betweenvariable settings of the standard, gated, and localized MEmodels with more conventional NN models.