The idea of replicating a common element in a more compact and universal version has been proposed as a solution to the replication needed in deep learning networks.