基于分步降维的高维学习索引研究  

Research on high-dimensional learned index based on stepwise dimensionality reduction

作　　者：刘进军 徐政 乔凯 方振益 LIU Jinjun;XU Zheng;QIAO Kai;FANG Zhengyi(College of Computer and Information Engineering,Hubei Normal University,Huangshi 435000,China)

机构地区：[1]湖北师范大学计算机与信息工程学院,湖北黄石435000

出　　处：《湖北师范大学学报（自然科学版）》2024年第3期20-24,共5页Journal of Hubei Normal University：Natural Science

摘　　要：在数据量和复杂性不断增加的时代,文本、音频和图像等高维数据的数量显著增长,这些数据的利用也变得更加频繁。因此,设计和实现一种高效的高维索引结构变得至关重要。基于降维的索引已被证明可以提高高维数据的查询效率。然而,随着数据量的增加,这些技术不可避免地会遇到诸如查询效率降低和内存使用增加之类的问题。为解决此问题,提出一种基于降维的高维学习索引,通过分步降维的方式,将高维数据降维为有序一维数据,并以此训练学习索引模型。在合成和真实数据集上的几个实验表明,该索引结构可以有效地提升查询效率及减少内存占用。In an era of dramatic increase in data volume and complexity,the amount of high-dimensional data such as text,audio,and images has grown significantly,and the utilization of these data has become more frequent.Therefore,it is of great importance to design and implement an efficient high-dimensional index structure.Studies has been proved that indexes based on dimensionality reduction can improve the query efficiency of high-dimensional data.However,as the amount of data increases,it is inevitable for these techniques to face challenges such as reduced query efficiency and increased memory usage.To solve this problem,this paper proposes a high-dimensional learned index based on dimensionality reduction,which reduces the dimensionality of high-dimensional data to ordered one-dimensional data by stepwise dimensionality reduction,and then trains the learned index.Experiments on synthetic and real-life datasets show that this index structure can effectively improve query efficiency and reduce memory consumption.

关 键 词：机器学习 降维 学习索引 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]