一种基于测地线的保局投影流形学习方法  

A method of manifold learning for Locality Preserving Projections based on geodesic

作　　者：徐礼俊 方婧涵 王亦平[1] Xu Lijun;Fang Jinghan;Wang Yiping(Navy Factory 902,shanghai 200083,China)

机构地区：[1]海军902厂

出　　处：《系统仿真学报》2019年第12期2892-2900,共9页Journal of System Simulation

摘　　要：为了解决在实际应用中LPP算法存在欠拟合状态的问题,详细论述了保局投影(LPP)的映射原理;分析了LPP方法在某些数据集下的欠拟合状态与邻接图之间的关系;提出了基于测地线的LPP(ISOLPP)流形学习方法。实验结果显示,ISOLPP方法在多个测试数据集上取得了很好的嵌入效果,不仅能够继承LPP具有显式投影矩阵的优点,而且解决了LPP算法中存在欠拟合状态的问题,显著提高了算法的适应性。In order to solve the problem of under-fitting state of LPP algorithm in practical application,in this paper, the mapping principle of Locality Preserving Projections(LPP) is discussed in detail. The relationship of LPP method between the under-fitting state on certain dataset and adjacency graph is analyzed. The LPP manifold learning method(ISOLPP) is proposed on the basis of geodesic. The experiment results show that the good embedded effect is achieved by implenmenting ISOLPP method on multiple test data sets. It significantly improves the adaptability of the algorithm by not only inheriting the advantages of LPP algorithm with explicit projection matrix, but also solving the disadvantages of LPP algorithm in the under-fitting state.

关 键 词：保局投影 欠拟合状态 测地线 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]