Matt Grover - 2002-09-29

Logged In: YES
user_id=171453

A backpropagation algorithm like this should fit into the
Amygdala design without too much trouble. Some changes will
have to be made to the way synapses are represented, but
those were already planned for the most part. It will also
require either separate Neuron classes (to override
Neuron::Train()), or a method for setting user-defined
training methods will have to be designed.

We will put this on the list of features to implement, but
it will probably be a few months before we can get to it
(although most of the design changes that are needed will be
done much earlier, so it would just be a matter of writing
the actual training function).