Concept Learning through Deep Reinforcement Learning with Memory-Augmented Neural Networks