21 Pages Posted: 17 Sep 2013 Last revised: 15 Jul 2014
Date Written: July 1, 2014
We study a robust model of the multi-armed bandit (MAB) problem in which the transition probabilities are ambiguous and belong to subsets of the probability simplex. We characterize the optimal policy as a project-by-project retirement policy but we show that arms become dependent so the Gittins index is not optimal. We propose a Lagrangian index policy that is computationally equivalent to evaluating the indices of a non-robust MAB. For a project selection problem we find that it performs near optimal.
Suggested Citation: Suggested Citation