Review for NeurIPS paper: Learning from Label Proportions: A Mutual Contamination Framework