I was reading about NN and they seemed very interesting and usefull. I was thinking about trying to make one myself but eaven after reading several articles I still don't really understand how they work. I think that my main problem is with the "hidden nodes", what do they do? I mean, If we know what output we want to get for a given input then why isn't the function enough? Also, how does the structure of the NN allow it to remember and optimize? Thanks.

Take a look in my journal way in the back. There is a small NN tutorial using a perceptron posted 7/1/2005. Maybe it will help?

Thanks, your tutorial was great! I understand now how the perceptron works, but that is with only one hidden node. I still don't get how it helps to have many of them and how you train it.
Thanks.

Read this topic and maybe you will get the idea of why we use hidden nodes.

You train a neural network (a perceptron anyway) by presenting the NN a input and a target output. The neural network generates an output from the input you presented. Then it generates an error (target output - output). This error is fed back to adjust weights internal to the neural network.