Question about MATLAB NEURAL NETWORK

Hi all,

Is there a way to figure out or know, if the trained network is in local minimum or global minimum. Or, if not, then as a designer do we just have the only option of playing around with different values for learning rate and momentum to see which actually gives best results out. Isn't there any way in MATLAB by which the local minimum and gloabl minimum can be observed ?

Also in relation to this context, reading documents, it's been stated in lot of them actually, that momentum helps in avoiding the network oscillate when large values for learning rate are chosen and also avoids the trap of the network falling into local minimum, I understand this concept, but just wanted to know then, how do we decide what value for momentum would be appropiate, Is this also like one of those trial and error RULES to see which value works best, or is there actually a way to figure out ?

Help would really be appreciated
Regards

In fact for any local learning algorithm like backpropagation, the minima will always be local.

for adjusting the learning rate and momuntm consider this rule if the learning error increases multiply the learning rate by .7 and set mom to 0 if the learning error decreases multiply the learning rate by 1.1 and set momentum to about .8.

for more information see matlab help.

also this is a very usful paper