Review for NeurIPS paper: Towards Understanding Hierarchical Learning: Benefits of Neural Representations