Differential Privacy in Deep Learning Models