Memorization in Deep Neural Networks: Does the Loss Function matter?