When you minimize loss using gradient descent, you are actually constantly negotiating with the limits of the information ...