Efficient Bayes-Adaptive Reinforcement Learning using Sample-Based Search