Markov Decision Processes In Deep Learning