Learning invariance through imitation