Abstract:
A multiple-alternative generalization of the familiar two-armed bandit problem is considered. A finite-state automaton that is asymptotically optimal in terms of complexity and that can solve the hypothesis-discrimination problem is created.