Learning processes in an asymmetric threshold network