基于图的无监督降维算法研究进展  

Research Progress in Graph-based Unsupervised Dimensionality Reduction Algorithms

作　　者：赵刚 刘飞飞 王继奎 聂飞平 ZHAO Gang;LIU Feifei;WANG Jikui;NIE Feiping(Gansu Provincial Institute of Natural Resources Planning and Research,Lanzhou 730000,China;School of Information Engineering and Artificial Intelligence,Lanzhou University of Finance and Economics,Lanzhou 730000,China;School of Artificial Intelligence,Optics and Electronics,Northwestern Polytechnical University,Xi'an 710000,China)

机构地区：[1]甘肃省自然资源规划研究院,甘肃兰州730000 [2]兰州财经大学信息工程与人工智能学院,甘肃兰州730000 [3]西北工业大学光电与智能研究院,陕西西安710000

出　　处：《信息与控制》2025年第1期28-49,共22页Information and Control

基　　金：国家自然科学基金项目(62176212);甘肃省自然科学基金项目(24JRRA1169)。

摘　　要：基于图的无监督降维是当前机器学习领域的研究热点,本文主要介绍了典型的基于图的无监督降维算法及其研究进展。首先,介绍了图的定义及其构造方法。然后,从4个方向介绍了12种经典及前沿方法,包括图固定的无监督降维算法、图固定的快速无监督降维算法、图优化的无监督降维算法和图优化的快速无监督降维算法,并对其进行了分析和总结。最后,对基于图的无监督降维技术的未来研究方向进行展望和总结。Graph-based unsupervised dimensionality reduction has become a research hotspot in machine learning.We review typical algorithms in this area and their research progress.We begin by defining the graphs and explaining their construction methods.Next,12 classical and cutting-edge methods are introduced,categorized into four groups:graph-fixed unsupervised dimensionality reduction algorithms,graph-fixed fast unsupervised dimensionality reduction algorithms,graph-optimized unsupervised dimensionality reduction algorithms,and fast unsupervised dimensionality reduction algorithms based on graph optimization.We then analyze and summarize these methods.Finally,future research directions for graph-based unsupervised dimensionality reduction techniques are discussed.

关 键 词：无监督学习 降维 图优化 时间复杂度 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]