Neural Network Toolbox

Summary

Single-layer linear networks can perform linear function approximation or pattern association.

Single-layer linear networks can be designed directly or trained with the Widrow-Hoff rule to find a minimum error solution. In addition, linear networks can be trained adaptively allowing the network to track changes in its environment.

The design of a single-layer linear network is constrained completely by the problem to be solved. The number of network inputs and the number of neurons in the layer are determined by the number of inputs and outputs required by the problem.

Multiple layers in a linear network do not result in a more powerful network, so the single layer is not a limitation. However, linear networks can solve only linear problems.

Nonlinear relationships between inputs and targets cannot be represented exactly by a linear network. The networks discussed in this chapter make a linear approximation with the minimum sum-squared error.

If the relationship between inputs and targets is linear or a linear approximation is desired, then linear networks are made for the job. Otherwise, backpropagation may be a good alternative.

Limitations and Cautions

Figures and Equations

© 1994-2005 The MathWorks, Inc.