Approximate nearest-neighbor search (ANNS), which is used to identify a small set of data points that are highly similar to a dynamically generated query, is widely used for applications ranging from recommendation systems to search engines. In some settings, ANNS must operate on data that cannot be stored completely in memory. To address this context, Coleman et al. present an approach (based on compressed sensing techniques and modern sketching algorithms) to compress the dataset into a sketch that can still identify near-neighbors in sublinear memory with high probability. They validate their approach with experiments on large social network datasets.