So can't you just use how real neurons learn as training data to to learn how to learn the same way?