15.04.2007 19:57, Guy Macon:

I see no reason why a neural network needs to be introduced
-- unless of course you want to investigate neural networks
as opposed to genetic algorithms.

No, not opposed. They work together. Fogel used an evolutionary
algorithm to optimize the weights of the neural network.

To investigate genetic
algorithms, take an existing, strong program and make many
copies with mutations of the evaluation functions, then let
them play against each other with winners having more
offspring.

Well, you have to parametrize somehow the evaluation function for this
approach. A neural network offers a very generic way to do so. What
parametrization would you suggest instead?

Greetings,
Ralf