Reasoning with deep learning: An open challenge