There's also a directed random search method you could use to train a neural network. For more information read this.

i'm testing the backpropagation method on a multi-layers net: two inputs, four hidden units and one ouput unit. when could i stop the training ?


i have calculated all error values and other stuffs, but the algorithm of my book does not tell me when we can stop the training....

If you're looking at human behaviours (i.e., realistic behaviours) in the FPS genre, then you absolutely must read Emma Norling's research. She covered this topic in her PhD dissertation several years back. You can find her research bibliography from this page.

Cheers,

Timkin

it seems that the dissertation has not yet been released on the net...