I was wondering what some ways to train NN are. I know that you can use Genetic Algorithms, back-propogation or self-train them, but what are some other ways? A Mathematical game developed and played excessively by friends and I

quote:

---

Original post by darkawakenings
I was wondering what some ways to train NN are. I know that you can use Genetic Algorithms, back-propogation or self-train them, but what are some other ways?

---

At its essence, training a neural network is an optimization problem: minimize expected error, given such-and-such weights to adjust. Therefore, any optimization technique which deals with continuous parameters might be used, such as simplex optimization, simulated annealing, etc. If you have a reasonably good training algorithm for your FFNN, though, I''d suggest investigating: 1. alternative neural architectures (and even- gasp!- non-neural modeling algorithms, like discriminant analysis, local regression, tree induction or k-nearest neighbors), which offer a variety of design trade-offs, and 2. data preparation tools, which can make an enormous difference in model performance.

good luck,
Predictor
http://will.dwinnell.com