On the value of learning for Bernoulli bandits with unknown parameters | IEEE Conference Publication | IEEE Xplore