基于可信赖性和连续性的流形降维效果评价方法  

Evaluation method for manifold dimensionality reduction effect based on trustworthiness and continuity

作　　者：刘丽娜[1,2] 马世伟[1] 芮玲 Liu Lina;Ma Shiwei;Rui Ling(School of Mechatronic Engineering ＆ Automation,Shanghai University,Shanghai 210072,China;School of Electrical ＆ Electronic Engineering,Shandong University of Technology,Zibo Shandong 255049,China)

机构地区：[1]上海大学机电工程与自动化学院,上海210072 [2]山东理工大学电气与电子工程学院,山东淄博255049

出　　处：《计算机应用研究》2018年第6期1707-1711,共5页Application Research of Computers

基　　金：国家自然科学基金资助项目(61671285;61601266);山东省自然科学基金资助项目(ZR2016FP04)

摘　　要：针对采用主观分析法对基于流形学习的非线性降维效果进行评价存在主观性强,缺乏必要的量化计算进行指导问题,提出利用可信赖性和连续性两个指标对流形降维效果进行量化评价。其中,可信赖性用于衡量流形降维可视化效果图的可信度,连续性则旨在分析原邻域的保持性。对常用的基于流形学习的非线性降维方法进行分类和对比研究,并在经典数据集Swissroll、Swisshole、Twopeaks、Helix和Puncturedsphere上利用可信赖性和连续性指标进行实验和对比分析,实验结果验证了该方法的有效性。In order to solve the problems of subjective error and lacking necessary quantify calculation for guidance by using subjective analysis method to evaluate the nonlinear data dimensionality reduction methods,this paper proposed an evaluative method based on two quantitative indicators to analysis the dimensionality reduction results. They were trustworthiness and continuity,the former aimed to quantify the trustworthiness of visual effect diagram for the manifold dimensionality reduction results,and the latter mainly analysed the original neighborhood preserving property. It classified and compared the existing nonlinear dimensionality reduction methods based on manifold learning. Meanwhile,based on trustworthiness and continuity,it implemented comparative experiments and analysis for some classical artificial datasets,such as Swissroll,Swisshole,Twopeaks,Helix and Puncturedsphere. The experimental results verify the efficiency of the proposed method.

关 键 词：数据降维 量化评价 可信赖性 连续性 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]