Multi-Level Spherical Locality Sensitive Hashing For Approximate Near Neighbors

Abstract

This paper introduces "Multi-Level Spherical LSH": parameter-free, a multi-level, data-dependant Locality Sensitive Hashing data structure for solving the Approximate Near Neighbors Problem (ANN). This data structure uses a modified version of a multi-probe adaptive querying algorithm, with the potential of achieving a $O(n^p + t)$ query run time, for all inputs n where $t <= n$.