Accelerating Large-Scale Inference with Anisotropic Vector Quantization

By Ruiqi Guo et al.

Read the original document by opening this link in a new tab.

Table of Contents

1 Introduction
2 Background and Related Works
3 Problem Formulation
4 Application to Quantization Techniques

Summary

Quantization based techniques are the current state-of-the-art for scaling maximum inner product search to massive databases. This paper introduces a family of anisotropic quantization loss functions that lead to a new variant of vector quantization. The proposed approach achieves state-of-the-art results on public benchmarks. The paper discusses the importance of efficient maximum inner product search (MIPS) and methods for accelerating MIPS, such as space partitioning and quantization. It also presents the anisotropic vector quantization problem and an iterative algorithm to optimize it.