A system theoretic perspective of learning and optimization

Learning and optimization of stochastic systems is a multi-disciplinary area that attracts wide attentions from researchers in control systems, operations research and computer science. Areas such as perturbation analysis (PA), Markov decision process (MDP), and reinforcement learning (RL) share the common goal. In this paper, we offer an overview of the area of learning and optimization from a system theoretic perspective. We show how these seemly different disciplines are closely related, how one topic leads to the others, and how this perspective may lead to new research topics and new results, and how the performance sensitivity formulas can serve as the basis for learning and optimization.