Maximum Principle Based Algorithms for Deep Learning