几种典型数据类型的谱聚类方法研究

主要针对几种典型数据的多流形结构分析问题进行了研究.综合分析多种谱聚类算法优缺点,以谱多流形聚类算法为主线,结合实验结果对多种谱聚类算法进行了分析,最后针对数据空间密度不均匀的情况对谱多流形聚类算法进行了一定的改进,提出了一种基于自适应近邻值的谱多流形聚类算法,并通过实验证明其达到了混合多流形聚类的目的.     

基于SMMC模型的数据多流形结构分析研究

采用混合多流形谱聚类模型(SMMC)对独立子空间、非独立子空间,非线性良分离及非线性交叉等流形聚类中的四种典型数据进行聚类,并与其他流形聚类方法进行比较,发现SMMC模型聚类效果良好且具有强鲁棒性和泛化能力.将SMMC模型运用于具有混合多流形结构的工件外部边缘轮廓进行聚类,结果显示SMMC模型能够很好的将其分为三类.针对SMMC模型复杂度高、选取参数困难及运行时间长的问题,提出了基于模拟退火遗传算法SMMC模型,结果发现改进后的模型能够大大缩短运行时间.     

基于数据流形结构的聚类方法及其应用研究

随着信息社会的不断发展,人类已经进入了信息爆炸时代,海量的数据使数据处理变得繁琐复杂,因此如何对现有的高维数据降维、聚类,并在一定程度上消除高维数据中存在的噪声是解决该问题的关键.基于相关的理论知识采用先降维后聚类的步骤,把高维数据按照子空间结构和流形结构两种情况分类,运用稀疏子空间聚类、谱多流形聚类、K-manifolds方法进行建模求解,通过对各种方法的对比,得出谱多流形聚类方法运行速度快,聚类准确度高,是最具有一般性特征的模型.     

混合流形结构的子空间聚类研究

主要讨论了线性流形和多流形的相关性分析、聚类分析等基本问题,在假设高维数据模型为多个子空间混合模型的基础上,分析了原始数据的几何结构特征,对于线性流形聚类问题采用稀疏子空间聚类算法(SSC),对于多流形聚类问题采用混合流形聚类算法(SMMC).此外,还通过对原始数据进行数据重采样,达到降维的目的,更有效的提取空间几何特征量,达到更好的聚类效果.     

数据的多流形结构分析

建立了基于ALM和权值的LRR聚类改进模型,对高维数据进行分析,将其分为两个独立的子空间,并与传统k-means聚类模型进行对比,采用评价指标模型对聚类结果进行评价分析.提出的LRR聚类改进模型在正则项引入了权重系数w,可以更好地将扰动分开,求解结果及评价指标均有效地验证了其稳定性、精确度等性能均有所提升.建立了SMMC改进模型,对机器工件外部边缘轮廓进行分类.从求解结果可看出该模型非常适合用于处理混合多流形聚类问题,对于比较复杂的曲线有着很好的分类性能.按照数据预处理、数据建模分析、模型结果评价步骤,通过使用谱聚类分析和多流形学习方法,对所给出的高维数据进行分析和处理,并通过评价模型得出相应的评价指标,对数据的多流形结构进行了深入的研究和探讨.     

基于sc RNA-seq数据的单细胞非线性降维方法

单细胞转录组测序数据中蕴含着丰富的细胞异质性表达信息,但也包含大量的冗余信息.降维不仅可以提取单细胞转录组测序数据内部的本质结构,减少冗余和噪声造成的误差,还可以为细胞聚类、基因富集分析、细胞发育轨迹推断等提供重要依据.本文介绍了基于流形学习、非负矩阵分解以及深度学习的非线性降维方法及其在单细胞转录组测序数据中的应用.     

正交约束优化问题的一阶算法

带有正交约束的矩阵优化问题在材料计算、统计及数据分析等领域中有着广泛的应用.由于正交约束的可行域是Stiefel流形,一直以来流形上的优化方法是求解这一问题的主要方法.近年来,随着实际应用问题所要求的变量规模的扩大,传统的流形优化方法在计算上的劣势显现出来,而一些迭代简单、收敛快的新算法逐渐被提出.通过收缩方法、非收缩可行方法、不可行方法三个类别分别来介绍求解带有正交约束的矩阵优化问题的最新算法.通过分析这些方法的主要特性,以及应用问题的要求,对这类问题算法设计的研究进行了展望.     

线性流形上矩阵方程B^TXB=D的反对称解

该文讨论了两类线性流形上矩阵方程B^TXB=D的反对称解和反对称最佳逼近解存在的条件,给出了通解的一般表达式,同时解决了解对给定矩阵的唯一最佳逼近问题.     

一种基于密度敏感的自适应谱聚类算法

针对传统的谱聚类算法不适合处理多尺度问题,引入一种新的相似性度量—密度敏感的相似性度量,该度量可以放大不同高密度区域内数据点间距离,缩短同一高密度区域内数据点间距离,最终有效描述数据的实际聚类分布.本文引入特征间隙的概念,给出一种自动确定聚类数目的方法.数值实验验证本文所提的算法的可行性和有效性.     

线性流形上双反对称矩阵的最佳逼近

本文讨论了线性流形上用双反对称矩阵构造给定矩阵的最佳逼近问题,给出问题解的表达式,最后给出求最佳逼近解的数值方法与数值算例.     

Stein fillability and the realization of contact manifolds

There is an intrinsic notion of what it means for a contact manifold to be the smooth boundary of a Stein manifold. The same concept has another more extrinsic formulation, which is often used as a convenient working hypothesis. We give a simple proof that the two are equivalent. Moreover it is shown that, even though a border always exists, its germ is not unique; nevertheless the germ of the Dolbeault cohomology of any border is unique. We also point out that any Stein fillable compact contact -manifold has a geometric realization in via an embedding, or in via an immersion.

     

A Note on P—Sasakian Manifold

s51.IntroductionLetMbeadifferentiablemanifo1dofdimensionn,ifMadmitsa(l,l)-tensorfieldgl,aPoSitivedefiniteRiemannianmetricg,avectorfieldeandal-formVwhichsatisfythefollow-ingconditions.thensuchamanifoldMiscalledaPara-SasakianmanifoldorbrieflyaP-Sasakianmanifoldby7'.AdatiandK.Matsumoto[ljwhichareconsideredasspecialcaseofanalmostparacontactmanifoldintroducedbyI.Sato[2].WhereVdenotestheoperatorofcovariantdifferentiationwithrespecttothemetrictensorg'X(M)denotesthesetofdifferentiablevectorfiel…     

On Flat Complete Causal Lorentzian Manifolds

We describe up to finite coverings causal flat affine complete Lorentzian manifolds such that the past and the future of any point are closed near this point. We say that these manifolds are strictly causal. In particular, we prove that their fundamental groups are virtually abelian. In dimension 4, there is only one, up to a scaling factor, strictly causal manifold which is not globally hyperbolic. For a generic point of this manifold, either the past or the future is not closed and contains a lightlike straight line     

ON PSEUDOSYMMETRY TYPE HYPERSURFACES OF SEMI-EUCLIDEAN SPACES I

In this paper the authors investigate hypersurfaces M of a semi-Euclidean space Esn+1, n > 4, satisfying (aC + BR) ?H = LkQ(g, Hk), k = 1,2, 3. Using obtained results they show additional curvature properties of investigated hypersurfaces.     

Some remarks about 1-convex manifolds on which all holomorphic line bundles are trivial

We give an abridged proof of an example already considered in [M. Col?oiu, On 1-convex manifolds with 1-dimensional exceptional set, Rev. Roumaine Math. Pures et Appl. 43 (1998) 97-104] of a 1-convex manifold X of dimension 3 such that all holomorphic line bundles on X are trivial. We also point out several mistakes of [Vo Van Tan, On the quasiprojectivity of compactifiable strongly pseudoconvex manifolds, Bull. Sci. Math. 129 (2005) 501-522] concerning this topic.     

ON PSEUDOSYMMETRY TYPE HYPERSURFACES OF SEMI-EUCLIDEAN SPACES Ⅰ

In this paper the authors investigate hypersurfaces M of a semi-Euclidean space Esn+1, n ≥ 4, satisfying (aC +βR). H ＝ LkQ(g, Hk), k ＝ 1,2,3. Using obtained results they show additional curvature properties of investigated hypersurfaces.     

Contactly geodesic transformations of almost-contact metric structures

We introduce the notion of contactly geodesic transformation of the metric of an almost-contact metric structure as a contact analog of holomorphically geodesic transformations of the metric of an almost-Hermitian structure. A series of invariants of such transformations is obtained. We prove that such transformations preserve the normality property of an almost-contact metric structure. We prove that cosymplectic and Sasakian manifolds, as well as Kenmotsu manifolds, do not admit nontrivial contactly geodesic transformations of the metric, which is a contact analog of the well-known result for Kählerian manifolds due to Westlake and Yano.     

The signature of generalized flag manifolds and applications

Let G be a linear algebraic group over C and P be a parabolic subgroup. We determine the signatures of the flag manifold G/P. As an application, we prove that the nonsingular hypersurfaces of degree 2 in CP^n are prime if n satisfies certain conditions.     

Invariant manifolds for a singular ordinary differential equation

We study the singular ordinary differential equation

(0.1)