基于核矩阵的Isomap增量学习算法研究.pdf

ISSN 1卜1777／TP 计算机研究与发展 1000—1239／CN of Journal Researchand Computer Development 基于核矩阵的Isomap增量学习算法研究 王耀南1 张 莹1 李春生1’2 1(湖南大学电气与信息工程学院长沙410082) 2(广东商学院数学与计算科学系广州 510320) (yaonan@hnu．on) KernelMatrixBasedIncremental LearningIsomapAlgorithm Li WangYaonanl，ZhangYin91，andChunshen91·2 Electricaland 410082) 1(Collegeof InformationEngineering，HunanUniversity，Changsha Mathematics8L Business 510320) 2(Department Studies，Guangzhou of ComputerScience。GuangdongUniversityof Abstract in thatalldatashouldtobeavailable The batch Isomapalgorithmoperates mode，meaning when is scenariosthedatacome andtheeffectofthe done．However，in sequentially training many dataisaccumulated．Whennewdatais takesa timetO the distance added．it long updategeodesic kernel incremental matrix alldata ordertO the based includingpoints．In improvecomputingspeed，a distancematrixcanbe as learningIsomapalgorithm(ILIsomap)isproposed．Thegeodesic interpreted akernel then a methodtO thatthe matrix，and guaranteegeodesic II。Isomapexploitsconstant—shifting distancematrix theMercer needstocalculatethe distance satisfy kernel．ILIsomaponly geodesic betweennewdataandthe