Bayesian optimal control of Markovian genetic regulatory networks

Finding optimal control policies for Markovian genetic regulatory networks requires that the transition probabilities be known precisely. In practice, due to practical limitations and complex nature of the underlying system, this knowledge may be inaccurate or incomplete. To address this difficulty, we construct an uncertainty set around the true network and define a probability distribution over this set, which reflects our prior knowledge about the underlying system. We take a Bayesian approach and formulate the optimal control policy minimizing the expected infinite-horizon discounted cost relative to this uncertainty class, resulting in an intrinsically robust policy.