An Optimal Computing Budget Allocation Tree Policy for Monte Carlo Tree Search

We analyze a tree search problem with an underlying Markov decision process, in which the goal is to identify best action at root that achieves highest cumulative reward. present new policy optimally allocates limited computing budget maximize lower bound on probability of correctly selecting each node. Compared widely used upper confidence (UCB) policies, presents more balanced approach manage exploration and exploitation tradeoff when sampling limited. Furthermore, UCB assumes support reward distribution known, whereas our algorithm relaxes this assumption. Numerical experiments demonstrate efficiency root.