Review for NeurIPS paper: What shapes feature representations? Exploring datasets, architectures, and training