Spreading vectors for similarity search

TL;DR

This paper introduces a novel approach to similarity search by training neural networks to adapt data to fixed, parameter-free quantizers, improving performance and flexibility in high-dimensional indexing.

Contribution

It proposes reversing the traditional quantizer training paradigm by adapting data to a fixed quantizer using neural networks with a new uniformity regularizer.

Findings

Outperforms most learned quantization methods

Competitive with state-of-the-art benchmarks

Training without quantization maintains accuracy

Abstract

Discretizing multi-dimensional data distributions is a fundamental step of modern indexing methods. State-of-the-art techniques learn parameters of quantizers on training data for optimal performance, thus adapting quantizers to the data. In this work, we propose to reverse this paradigm and adapt the data to the quantizer: we train a neural net which last layer forms a fixed parameter-free quantizer, such as pre-defined points of a hyper-sphere. As a proxy objective, we design and train a neural network that favors uniformity in the spherical latent space, while preserving the neighborhood structure after the mapping. We propose a new regularizer derived from the Kozachenko--Leonenko differential entropy estimator to enforce uniformity and combine it with a locality-aware triplet loss. Experiments show that our end-to-end approach outperforms most learned quantization methods, and is competitive with the state of the art on widely adopted benchmarks. Furthermore, we show that training without the quantization step results in almost no difference in accuracy, but yields a generic catalyzer that can be applied with any subsequent quantizer.