Monitoring with Limited Information

We consider a system with an evolving state that can be stopped at any time by decision maker (DM), yielding state-dependent reward. The DM does not observe the except for limited number of monitoring times, which he must choose, in conjunction suitable stopping policy, to maximize his Dealing these types problems, arise variety applications from healthcare finance, often requires excessive amounts data calibration purposes and prohibitive computational resources. To overcome challenges, we propose robust optimization approach, whereby adaptive uncertainty sets capture information acquired through monitoring. two versions problem—static dynamic—depending on how times are chosen. show that, under certain conditions, same worst-case reward is achievable either static or dynamic This allows recovering optimal policy resolving problem. discuss cases when problem becomes tractable highlight conditions equidistant optimal. Lastly, showcase our framework context (monitoring heart-transplant patients cardiac allograft vasculopathy), where design policies substantially improve over status quo recommendations. paper was accepted Chung Piaw Teo, optimization.