一种新有监督流形学习方法.pdfVIP

计算机研究与发展 1SSN1000一1239fCN1l—1777／TP of Re鲫rchand J伽malComputer Developnlent 一种新的有监督流形学习方法 孟德宇1‘2徐宗本1戴明伟2 1(西安交通大学信息与系统科学研究所西安71呻49) 2(西安交通大学电子与信息工程学院西安710049) (dym耶g@mail．卅tu．edu．cn) ANew Manifold Method SuperVised Learning Dai Me“gDeyul一，XuZongbenl，andMingwei2 1(，nmfn据归I咖删￡f删ndS弦招辨蝌姘，崩’dnJ缸帕埘耳unz唧m_，刑’n”710049) 2(＆^。。f0，Ektm缸d耐f”，or帆n加nE”一月料i”g，崩’nnJE口咖ngmf钟m廿，弼’删710049) Abst憎ctAnew manifold methodis inthis orderto anew supervised 1earning proposedpaper，in present to andr10niinear to effIcientlyman；fold reductionmethods strategy apply Iearnj“g dimensionality supervised newmethodrealizesefficient basedon the leamingprobIems．The supervisedlearningmainly integratirlg ofthemanifo【d and some topoIogypreseⅣingproperty Iearningmethods(IsomapU。E)andprominent of vectormachine onmiddIeand锄ansizeddatasetsand既sential properties s“pport such船effIcjency of vecto碍calculatedfrom vectormachine．Themethodisrealizedviathe capabilitysupport support to orLLE the to ofthe data跎tinthe10w followingsteps：fjrstapplyIsomap get embeddi“gsorlginal hlcharethe dImensional toobtain most andintrinsicdataforthe space；then supportvectors，w significant finaIcIassification vectormachineonthese【owdjmensional resuIt，byusingsupport embeddingdata； t0 vectorsinthe dim朗sionalbased0nthe 1abels support