Word2Bits - Quantized Word Vectors
This addresses memory constraints for resource-limited devices like mobile phones and GPUs, offering an incremental improvement in efficiency and performance.
The paper tackles the problem of high memory and storage requirements for word vectors by learning quantized versions using 1-2 bits per parameter, which reduces space by 8-16x and outperforms full-precision vectors on word similarity and question answering tasks.
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.