A Tour of Locality Sensitive Filtering on the Sphere

TL;DR

This paper provides a unified overview and analysis of Locality Sensitive Filtering techniques for the Angular Approximate Near Neighbor problem, highlighting their connections, optimality, and core mechanisms.

Contribution

It introduces a new LSF-based data structure for Angular ANN and offers a streamlined, optimal analysis connecting existing techniques and results.

Findings

Designed and analyzed an LSF-based data structure for Angular ANN

Proved the optimality of the proposed data structure

Revisited and strengthened a key technical lemma

Abstract

The Approximate Near Neighbor (ANN) problem is a cornerstone in high-dimensional data analysis, with applications ranging from information retrieval to data mining. Among the most successful paradigms for solving ANN in high-dimensional metric spaces is Locality Sensitive Hashing (LSH), alongside the more recent Locality Sensitive Filtering (LSF). Since the seminal work of Indyk and Motwani, literature has expanded into a complex landscape of variants, often presented under different perspectives and adopting different notation. In this work, we address the technical challenge of navigating this landscape, by providing a self-contained, unified view of the essential algorithmic ingredients governing LSH-based and LSF-based solutions for angular distance -- a case of particular relevance in modern applications. In doing so, we touch on deep connections between LSF and LSH strategies. Our contribution is twofold. First, we design and analyze an LSF-based data structure for the Angular ANN problem, serving as a "guided tour" through fundamental techniques and results in the area. Second, we provide a streamlined analysis that, piecing together technical ingredients and results appeared throughout previous literature, proves optimality of the proposed data structure. In doing so, we revisit and strengthen a key technical lemma central to this body of work. The result is a critical and cohesive review that identifies core mechanisms of proximity search, offering both a streamlined entry point for researchers and a refined perspective on the state of the art.