Which neural network toolbox can help me getting exact feedback parameters?

Hi!

First of all, I'm not mainly an IT guy, so i want to construct my network with a toolbox instead of using codes, if I can. My subject is, i have to make a network about mechanical elaboration, i have 3d scans of the final surface , and also the processing parameters of each surface. I have to teach the network for what each elabortion parameter change cause to the end product, so if i give an input matrix of an unknown surface to the ready network, it will tell me the processing parameters, even if it's not completly similar to the ones which i teached it.

I already tried pattern recognition toolbox, but it can only classify the surfaces in binary way. Now I'm trying self-organizing map, but I would really appreciate if you would help me, which way is right, or less if i were able to do this, or not. This project is important for me, and the deadline is getting really close.

Thank you so much for every answer, and sorry for my silly english.

David

Answer 1

[Greg Heath]

It is not true that patternet can only classify exclusive classes.

If, instead of {0,1} unit vector target columns, additional ones are added to indicate co-occurring classes like [ short, fat, ugly, smelly], more than a single one can be used in the target column.

Thresholds can then be used on the output to decide which characteristics are present.

If interested, a Bayesian approach can be implemented using a priori probabilities and misclassification costs.

Hope this helps.

Greg