When applying pre-trained networks to a new dataset, attention models can be used to determine how much weight to assign to the input of each network