Loss Function In Supervised Learning