Accelerated learning in back-propagation nets