Review for NeurIPS paper: HM-ANN: Efficient Billion-Point Nearest Neighbor Search on Heterogeneous Memory

The paper attempts to scale nearest neighbor search using heterogenous memory hardware. In this regard, authors devised a practical trick on top of HNSW. It is a clean node promotion strategy along the memory hierarchy using the degree information. The method was evaluated on some common large datasets, but not necessarily difficult ones. Reviewers found the setup to leverage the memory hierarchy interesting and the benefits obtained from it appears promising.