Loss Function In Deep Learning