Learning to compute inner consensus -- A noble approach to modeling agreement between Capsules