Review for NeurIPS paper: Compositional Visual Generation with Energy Based Models