What is the normalization parameter for when using cross-entropy as a performance function of a patternnet?

I'm currently investigating the possiblities of using the Matlab function patternnet to generate and train a pattern recognition ANN.

Within the performance function used here (cross-entropy), there is an option to set network.performParam.normalization = 'standard'.

My question is: What is the benefit of normalizing targets / outputs by setting this option, when the target vectors should be designed as [000010000] anyways?

So the range of the outputs of the individual output neurons and the targets is 0...1 anyway?

Thanks!

Answer 1

[Sahithi Kanumarlapudi]

‘performParam.normalization’ parameter is designed to be used with any neural network. Setting its value to ‘standard’ results in outputs and targets being normalized to (-1, +1), and therefore errors in the range (-2, +2) which is helpful in many cases.

But in case of pattern recognition networks this might not be helpful as the outputs and targets are already (0,1).