基于一致互补性的多视角最小二乘支持向量机  

Consensus and complementarity-based multi-view least square support vector machine

作　　者：唐静静 李佳辉 田英杰[3,4] TANG Jingjing;LI Jiahui;TIAN Yingjie(School of Business Administration,Southwestern University of Finance and Economics,Chengdu 611130,China;Institute of Big Data,Southwestern University of Finance and Economics,Chengdu 611130,China;Research Center on Fictitious Economy and Data Science,Chinese Academy of Sciences,Beijing 100190,China;School of Economics and Management,University of Chinese Academy of Sciences,Beijing 100190,China)

机构地区：[1]西南财经大学工商管理学院,成都611130 [2]西南财经大学大数据研究院,成都611130 [3]中国科学院虚拟经济与数据科学研究中心,北京100190 [4]中国科学院大学经济与管理学院,北京100190

出　　处：《系统工程理论与实践》2022年第9期2461-2471,共11页Systems Engineering-Theory & Practice

基　　金：国家自然科学基金(71901179,71991472,12071458,71731009)。

摘　　要：多视角学习(multi-view learning)是指利用事物的多视角数据,对其内在模式进行识别和学习.然而,大部分多视角学习模型存在两个弊端:1)仅适用于两视角学习场景,对于视角个数超过两个的情形便无法直接进行处理;2)要么遵循一致性原则,要么遵循互补性原则,同时遵循两个原则的研究工作比较少.为解决以上两个弊端,本文通过使用最小二乘损失函数和权重分配策略,构建了基于一致性和互补性原则的多视角最小二乘支持向量机(multi-view least square support vector machine with the consensus and the complementarity principles,MVLSSVM-2C),并设计了相应的交替优化算法对模型进行求解.进一步地,本文利用Rademacher复杂度理论对MVLSSVM-2C的泛化能力进行分析.最后,在大量的多视角数据集上验证了MVLSSVM-2C模型的合理有效性.Multi-view learning(MVL)exploits the multi-view data to improve the performance of the learning tasks.However,most multi-view leaning models are built for only two-view setting,or mainly embed either the consensus principle or the complementarity principle.To overcome aforementioned drawbacks,we propose a consensus and complementarity-based multi-view least square support vector machine(MVLSSVM-2C),which leverages view-agreement on multi-view predictors and weight combination strategy.We then adopt an iterative two-step strategy to solve the optimization problem efficiently.Further more,the generalization capability is theoretically analyzed by using Rademacher complexity.The extensive experiments validate the effectiveness of the proposed model.

关 键 词：多视角学习 最小二乘支持向量机 一致性原则 互补性原则 

分 类 号：F270[经济管理—企业管理]