The Mixture of Neural Networks Adapted to Multilayer Feedforward Architecture

摘要

The Mixture of Neural Networks (MixNN) is a Multi-Net System based on the Modular Approach. The MixNN employs a neural network to weight the outputs of the expert networks. This method decompose the original problem into subproblems, and the final decision is taken with the information provided by the expert networks and the gating network. The neural networks used in MixNN are quite simple so we present a mixture of networks based on the Multilayer Feedforward architecure, called Mixture of Multilayer Feedforward (MixMF). Finally, we have performed a comparison among Simple Ensemble, MixNN and MixMF. The methods have been tested with six databases from the UCI repository and the results show that MixMF is the best performing method.