Stochastic Control Approach To The Multi-Armed Bandit Problems