what if 

 Every neural network contained 

a smaller ,refined neural network within, 

trained exactly as per itself, 

but with attenuated signal levels

Therefore the internal neural network

Functions as a governor...?

also see : bagging and boosting in random forests