verification neural network application

Hello,

I run character recognition, neural network application in Matlab, after training the network I reach layer one neuron weights, I would like change some of weights approximately and check how much difference between results of original weights and changed weights?

Is there any function to find percentage of verification? for example, after training network, simulate with specific input data and find output accuracy 98% then I would change some weights approximately and find percentage accuracy 90% I couldn't find any thing useful.

Any help would be greatly appreciated, Thanks.

Answer 1

[Greg Heath]

There is no function for doing this for a very good reason: It is a pretty useless exercise.

More important investigations:

 1. The minimum number of hidden nodes that yields acceptable results on non-training data

 2. For multiple input nets: 

     a. Importance ranking of inputs on non-training data
     b. Minimum number of inputs needed to obtbain acceptable results on non-training data

Hope this helps to keep things in perspective.

Greg

[Ali]

thank you very much for reply but I don't get my answer yet, I am new in Matlab, and just want compare result of a neural network application by its golden target in percentage,