On optimal control and machine learning