arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2401.14732cs.LG
收藏

基于隐式神经码本的残差量化

Residual Quantization with Implicit Neural Codebooks

FAIR at Meta(Meta FAIR实验室);Eindhoven University of Technology(埃因霍温理工大学)

Iris A. M. Huijben, Matthijs Douze, Matthew Muckley, Ruud J. G. van Sloun, Jakob Verbeek

AI总结:

针对传统残差量化未考虑误差与已选码字依赖的问题,提出基于前序近似构建步骤专用码本的神经残差量化方法QINCo,在多个数据集上以更小码长实现更优的最近邻搜索精度。

摘要:

向量量化是数据压缩与向量搜索中的一项基础操作。为实现高精度,多码本方法通过多个码本的码字来表示每个向量。残差量化(RQ)便是这类方法之一,它迭代量化前一步的误差。尽管误差分布与先前选定的码字相关,但传统残差量化在每个量化步骤均使用固定码本,并未考虑这种依赖关系。本文提出QINCo,这是一种神经残差量化变体,它能根据前几步得到的向量近似结果,为每一步构建专用码本。实验表明,在多个数据集和不同码长下,QINCo的性能大幅优于现有最优方法。例如,在BigANN1M和Deep1M数据集上,QINCo使用12字节编码的最近邻搜索精度,优于使用16字节编码的现有最优方法UNQ。

英文摘要:

Vector quantization is a fundamental operation for data compression and vector search. To obtain high accuracy, multi-codebook methods represent each vector using codewords across several codebooks. Residual quantization (RQ) is one such method, which iteratively quantizes the error of the previous step. While the error distribution is dependent on previously-selected codewords, this dependency is not accounted for in conventional RQ as it uses a fixed codebook per quantization step. In this paper, we propose QINCo, a neural RQ variant that constructs specialized codebooks per step that depend on the approximation of the vector from previous steps. Experiments show that QINCo outperforms state-of-the-art methods by a large margin on several datasets and code sizes. For example, QINCo achieves better nearest-neighbor search accuracy using 12-byte codes than the state-of-the-art UNQ using 16 bytes on the BigANN1M and Deep1M datasets.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑