[ODE] Community page

Mattias Fagerlund mattias at cambrianlabs.com
Thu Oct 10 10:10:02 2002


> > My current implementation evolve NNs unsing the NEAT method. Nothing 
> > online yet, though.
> 
> I've thought about NNs but i'm not really upto speed on them. I should
> really investigate the area more.

NNs are tricky, if you want to do any kind of "fancy" (backpropagation)  
learning. Evolving NNs is almost trivial. Picking the right network 
topology is the the tricky part. The method I use, NEAT, evolves 
topology AND network weights, that's harder to implement.

Now, if you want to _understand_ NNs, that's another thing completely...

cheers,
m