Competitive networks with recurrent connections

[Dennis Garlick <dennisg@psych.usyd.edu.au>]

Hi,

I'm interested in trying competitive networks that have recurrent 
connections. However, when I use the so++ module that has soft competitive 
learning, it seems as though the activation is progressively calculated 
from layer 0 onwards, and the effect of later layers isn't taken into 
account when calculating earlier layers. Another binary does recurrent 
connections through iterations, but it doesn't have the soft competitive 
learning algorithm. Is there a way that I can use both soft competitive and 
recurrent connections in the same network and have the activations 
calculated correctly?

Thanks for any help,

Dennis