Concept learning through deep reinforcement learning with memory-augmented neural networks