It's difficult to design meaningful SNN top down. If you
want to use the power of SNN the network got to be highly recurrent
and is therefore extremely unpredictable.
To overcome these limitations, a genetic algorithm can be employed.
The idea is that the shape (the topology) of the net translates into
fitness for a particular task.
The GA built into Amygdala is divided into a client and a server part
to allow for distributed operation. The server manages the genome
population while the client evaluates them. The protocol used to
download new genomes and submit scores is HTTP. It is even possible
to use vast populations by running the GA on the Internet.