How to improve artificial neural network performance

I want to know how to increase artificial neural network performance. Whether increment of the number of neurons will help me or not. Please help

Answer 1

[Walter Roberson]

If you have too many neurons, you can get over-training -- cases where the NN recognizes specific input data rather than recognizing patterns.

If you do not have enough layers, some types of recognition might not be possible. I have not examined NN enough to know what might happen if you have too many layers.

What kind of recognition are you trying to do?

[DIBYENDU SUR]

ionosphere data recognition, huge data set, 207697 data per parameter

[DIBYENDU SUR]

one data ranges from 1 to 365, another 0 to 24, another 20 to 24, another 85 to 91, and the last input parameter is of 0 to 60, output data ranges from 0 to 50

[Greg Heath]

What, exactly, are the inputs and outputs?

Why is your current performance unsatisfactory?

Greg

[DIBYENDU SUR]

my error goal was set to 0, but it does not come under 15, i have used msereg instead of mse as the function, increases neurons, even tried by normalizing the inputs inside 0 to 1, tried with 2,3 layers, tried with 64 bit OS also, but the problem persists

[Greg Heath]

Not enough info. HOw about posting some code?

Greg