next up previous
Next: Training the System Up: A Modular Neural Previous: Introduction

The Network Architecture

The proposed network system consists of a layer of input modules and an additional decision module. All sub-networks are MLPs. Each input variable is connected to only one of the input modules. These connections are chosen at random. The outputs of all input modules are connected to the decision network. The structure is depicted in Figure 1.

The following parameters are assumed: the dimension of the input vector is l and the number of classes is k.

Figure 1:  The Multiple Neural Network Architecture.

One of the design issues is to select the number of inputs per module in the first layer (n); this decision determines the number of input modules tex2html_wrap_inline352. (It is assumed that l = m*n; if this is not the case the spare inputs may be connected to constant inputs or the size of one of the networks may be altered.) Each network in the first layer has tex2html_wrap_inline356 outputs. This is the required number to represent all the classes in a binary code.

The decision network has tex2html_wrap_inline358 inputs. The number of outputs is k, one neuron for each class.

The number of weights is much less than in a fully connected monolithic MLP with the same number of hidden neurons.

Albrecht Schmidt and Zuhair Bandar
Wed Apr 16 13:57:11 MET DST 1997