Learning, Control, and Reduction for Markov Jump Systems