No, but do that partially 100,000 times with weights overfitted to a sample dataset and you are :^)