Multi-Level Spherical Locality Sensitive Hashing For Approximate Near Neighbors

TL;DR

This paper presents Multi-Level Spherical LSH, a parameter-free, data-dependent hashing method for efficient approximate near neighbor searches, utilizing a multi-probe adaptive query algorithm to improve query times.

Contribution

It introduces a novel multi-level spherical LSH data structure that is parameter-free and data-dependent, enhancing the efficiency of approximate near neighbor searches.

Findings

Achieves a query runtime of approximately O(n^p + t) for all inputs.

Employs a modified multi-probe adaptive querying algorithm.

Provides a data-dependent, parameter-free LSH framework.

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 $t <= n$.