Word2Bits - Quantized Word Vectors

TL;DR

This paper introduces a method to create highly compressed word vectors using 1-2 bits per parameter, which are memory-efficient, act as regularizers, and outperform full-precision vectors on several NLP tasks.

Contribution

The paper presents a novel quantization approach integrated into Word2Vec, enabling high-quality, extremely compact word vectors that outperform traditional vectors on key NLP benchmarks.

Findings

Quantized vectors use 8-16x less space than full precision vectors.

Quantized vectors outperform full precision vectors on word similarity tasks.

Training with quantization acts as an effective regularizer.

Abstract

Word vectors require significant amounts of memory and storage, posing issues to resource limited devices like mobile phones and GPUs. We show that high quality quantized word vectors using 1-2 bits per parameter can be learned by introducing a quantization function into Word2Vec. We furthermore show that training with the quantization function acts as a regularizer. We train word vectors on English Wikipedia (2017) and evaluate them on standard word similarity and analogy tasks and on question answering (SQuAD). Our quantized word vectors not only take 8-16x less space than full precision (32 bit) word vectors but also outperform them on word similarity tasks and question answering.