Multi-armed bandit problem with intra- and inter- correlations

Abstract

We address a veiled expected reward multi-slot machine problem. Object of this problem is to select the highest expected reward arm which is called optimal arm. This problem is usually called Multi-armed Bandit problem which is famous for reinforcement learning problem about exploration and exploitation problem. In order to extend the MAB problem, we consider Unimodality correlation between arm and additionally observable conditions. In the proposed framework, we prove the asymptotic fundamental limit and suggest an algorithm which is achieving limit.