You reduce the resolution of the training data so your model doesn’t have to process as much info this means you have more resources to train other parts of your model