Approximate Nearest Neighbor Search on High Dimensional Data --- Experiments, Analyses, and Improvement (v1.0)

TL;DR

This paper provides a comprehensive evaluation of state-of-the-art approximate nearest neighbor search algorithms across various domains, datasets, and settings, and introduces a new method that improves efficiency and recall.

Contribution

It offers the first extensive cross-disciplinary analysis of ANNS algorithms and proposes a novel method that enhances both query efficiency and accuracy.

Findings

Most algorithms perform well on certain datasets but vary significantly across others.

The new method consistently outperforms existing algorithms in efficiency and recall.

Comprehensive evaluation reveals strengths and weaknesses of current ANNS approaches.

Abstract

Approximate Nearest neighbor search (ANNS) is fundamental and essential operation in applications from many domains, such as databases, machine learning, multimedia, and computer vision. Although many algorithms have been continuously proposed in the literature in the above domains each year, there is no comprehensive evaluation and analysis of their performances. In this paper, we conduct a comprehensive experimental evaluation of many state-of-the-art methods for approximate nearest neighbor search. Our study (1) is cross-disciplinary (i.e., including 16 algorithms in different domains, and from practitioners) and (2) has evaluated a diverse range of settings, including 20 datasets, several evaluation metrics, and different query workloads. The experimental results are carefully reported and analyzed to understand the performance results. Furthermore, we propose a new method that achieves both high query efficiency and high recall empirically on majority of the datasets under a wide range of settings.