一种稳健的聚类方法

本文讨论一种新的聚类方法 :属性均值聚类 .通过理论分析 ,属性均值聚类是比模糊均值聚类更稳健的聚类方法 .数值实验说明了该方法的有效性     

基于三支决策的灰色可能度聚类方法及应用

针对经典的灰色可能度聚类评估模型难以判定决策对象的灰类归属和过度聚类等问题,利用三支决策的思想和方法,通过引入三支灰类的概念描述决策对象和灰类之间的不确定聚类关系;将其代替灰色定权聚类中的灰类和严格的聚类关系,构建基于三支决策的灰色可能度聚类方法,并采用决策粗糙集中的贝叶斯推理确定聚类阈值;最后,以案例验证所提方法的有效性和合理性。结果表明：本文所构建的模型是经典灰色可能度聚类评估模型的拓展和泛化,可以有效避免过度聚类,降低决策风险,提高聚类可靠性。     

海外旅游流的网络相关性

本使用最小距离聚类法和最大树模糊聚类法对中国主要旅游热点城市海外旅游流的流动网络进行了分类，做了结点间的相关性分析。     

等混合聚类法的介绍

一、引言 聚类分析法是要解决在地质领域(或其它领域)中常碰到的分类问题.此方法在地质领域中的应用较为广泛.已用于地质上的地层对比、分类、石油地质上调整井油层水淹级划分及油气资源评价等方面,收到良好的地质效果. 目前我国常用的聚类方法,一般可分为系统聚类和动态聚类两种,但目前常用的聚类分析法,多变样本数量不能太大的限制.有的虽然能处理大数据量的资料,但往往花费的机时太长. 等混合聚类法是一种新的聚类法,这种方法不是一般的将一个样品和另外所有样品逐个加以比较,而是先将原始样品预分为若干组,然后根据一定原则,将各组分别…     

双向聚类方法综述

传统的聚类方法由于无法提取样本和变量间的局部对应关系,并且当数据具有高维性和稀疏性时表现不佳,因此学者们提出了双向聚类,基于样本和变量间的局部关系,同时对样本和变量进行聚类,形成一系列子矩阵的聚类结果。近年来,双向聚类发展迅速,在基因分析、文本聚类、推荐系统等领域应用广泛。首先,对双向聚类方法进行梳理与归纳,重点阐述稀疏双向聚类、谱双向聚类和信息双向聚类三类方法,分析它们之间的区别和联系,并且介绍这三类方法在多源数据的整合分析、多层聚类、半监督学习以及集成学习上的发展现状和趋势;其次,重点介绍双向聚类在基因分析、文本聚类、推荐系统等领域的应用研究情况;最后,结合大数据时代的数据特征和双向聚类存在的问题,展望双向聚类未来的研究方向。     

聚类问题的人工神经网络方法

在不同的实际问题中，往往视需要使用不同的准则对模式进行聚类。本文给出了一个聚类准则，并使用该准则用人工神经网络方法在计算机上进行了模拟。结果表明本文使用的聚类准则更适合于用人工神经网络实现，可以取得极好的聚类效果。     

系统聚类递推公式推广

本文推广了系统聚类法的最短距离法、最长距离法，平均距离法，中国距离法，重心法，类平均法、离差平方和法、可变法、可变类平均法的递推公式，并给出了统一计算公式。     

硬聚类和模糊聚类的结合——双层FCM快速算法

模糊c均值(FCM)聚类算法在模式识别领域中得到了广泛的应用,但FCM算法在大数据集的情况下需要大量的CPU时间,令用户感到十分不便,提高算法的速度是一个急待解决的问题。本文提出的双层FCM聚类算法是一种快速算法,它体现了硬聚类和模糊聚类的结合,以硬聚类的结果对模糊聚类的初始值进行指导,从而明显地缩短了迭代过程。双层FCM算法所用的CPU时间仅为FCM算法的十三分之一,因而具有很强的实用价值。     

函数型聚类分析:基于距离的一步法框架

基于距离的函数型聚类分析包含曲线拟合和聚类两个独立步骤,最优曲线拟合未必有利于类别信息的提取和保留。根据曲线拟合与聚类分析的计算过程,重新梳理了函数型聚类算法;基于距离度量,提出了同时考虑拟合和聚类效果的函数型聚类一步法;在交替方向乘子法(ADMM)框架下推导并给出了迭代求解算法。模拟试验结果显示,该函数型聚类算法有助于提高聚类精度;针对北京市空气质量监测站点二氧化氮(NO_2)污染物小时浓度数据的实例验证分析表明,该函数型聚类算法对不同类别空气质量监测点具有更好的区分度。     

区间模糊ISODATA动态聚类算法

目前模糊技术已经应用于许多智能系统,如模糊关系与模糊聚类.聚类是数据挖掘的重要任务,它将数据对像分成多个聚类,在同一个聚类中,对象的属性特征之间具有较高的相似度,有很大研究及应用价值.结合数据库中的挖掘技术,对属性特征为区间数的多属性决策问题,提出了一种基于区间数隶属度的区间模糊ISODATA动态聚类方法.     

A quantum analogue of generic bases for affine cluster algebras

We construct quantized versions of generic bases in quantum cluster algebras of finite and affine types.Under the specialization of q and coefficients to 1,these bases are generic bases of finite and affine cluster algebras.     

基于投入产出表的区域价值链集群识别研究——以上海为例

区域价值链集群分析是集群分析作为区域经济创新性和竞争优势分析工具的重要表现形式。而区域价值链集群识别是进行区域价值链集群分析的第一步。为了实现区域价值链集群依据经济系统各部门间网络互动联系来确定其核心主导产业及其前后向密切关联产业的识别目标,本文选取了以最大值法和投入产出分析法相结合的区域价值链识别方法。由此进行的识别研究表明,上海共有通信、化工、通专、金属、石油核燃料五类价值链集群,它们通过内部和彼此间的互动交织构成了密切关联的网络,并在很大程度上反映了上海经济系统的核心组成。而相应的经济技术分析表明,通专价值链集群将很有可能在上海大力发展先进制造业的产业结构升级大背景下赶超通信价值链集群而成为新一轮发展的亮点。     

Periodicities in Cluster Algebras and Cluster Automorphism Groups

We study the relations between two groups related to cluster automorphism groups which are defined by Assem,Schiffler and Shamchenko.We establish the relation-ships among (strict) direct cluster automorphism groups and those groups consisting of periodicities of labeled seeds and exchange matrices,respectively,in the language of short exact sequences.As an application,we characterize automorphism-finite cluster algebras in the cases of bipartite seeds or finite mutation type.Finally,we study the relation between the group Aut(A) for a cluster algebra A and the group AutMn(S) for a mutation group Mn and a labeled mutation class S,and we give a negative answer via counter-examples to King and Pressland's problem.     

REPETITIVE CLUSTER-TILTED ALGEBRAS

Let H be a finite-dimensional hereditary algebra over an algebraically closed field k and C F m be the repetitive cluster category of H with m ≥ 1. We investigate the properties of cluster tilting objects in C F m and the structure of repetitive clustertilted algebras. Moreover, we generalize Theorem 4.2 in [12] (Buan A, Marsh R, Reiten I. Cluster-tilted algebra, Trans. Amer. Math. Soc., 359(1)(2007), 323-332.) to the situation of C F m , and prove that the tilting graph KCFm of C F m is connected.     

The Cluster Function of Single-Valued Functions

Given a single-valued function f between topological spaces X and Y, we interpret the cluster set C(f;x) as a multivalued function F=C(f;⋅) associated to f – the cluster function of f. For appropriate metrizable spaces X and Y, we characterize cluster functions C(f;⋅) among arbitrary set-valued functions F and show that every cluster function F=C(f;⋅) admits a selection h of Baire class 2 such that F=C(h;⋅). Mathematics Subject Classifications (2000) Primary: 54C50, 54C60; secondary: 26A21, 54C65.This research was partially supported by DFG Grant RI 1087/2.     

Repetitive cluster-tilted algebras

Let H be a finite-dimensional hereditary algebra over an algebraically closed field k and C_Fm be the repetitive cluster category of H with m ≥ 1. We investigate the properties of cluster tilting objects in C_Fm and the structure of repetitive cluster-tilted algebras. Moreover, we generalize Theorem 4.2 in [12] (Buan A, Marsh R, Reiten I. Cluster-tilted algebra, Trans. Amer. Math. Soc., 359(1)(2007), 323-332.) to the situation of C_Fm, and prove that the tilting graph KC_Fm of C_Fm is connected.     

连续调查的整群抽样

本文讨论了连续调查中应用普查方法时的整群抽样和分层整群抽样。     

Applications of BGP-reflection functors: isomorphisms of cluster algebras

Given a symmetrizable generalized Cartan matrix A, for any index k, one can define an automorphism associated with A, of the field Q(u1,…, un) of rational functions of n independent indeterminates u1,…,un.It is an isomorphism between two cluster algebras associated to the matrix A (see sec. 4 for the precise meaning). When A is of finite type, these isomorphisms behave nicely; they are compatible with the BGP-reflection functors of cluster categories defined in a previous work if we identify the indecomposable objects in the categories with cluster variables of the corresponding cluster algebras, and they are also compatible with the "truncated simple reflections" defined by Fomin-Zelevinsky. Using the construction of preprojective or preinjective modules of hereditary algebras by DIab-Ringel and the Coxeter automorphisms (i.e. a product of these isomorphisms), we construct infinitely many cluster variables for cluster algebras of infinite type and all cluster variables for finite types.     

Preprojective Cluster Variables of Acyclic Cluster Algebras

It is proved that any cluster-tilted algebra defined in the cluster category 𝒞(H) has the same representation type as the initial hereditary algebra H. For any valued quiver (Γ, Ω), an injection from the subset 𝒫?(Ω) of the cluster category 𝒞(Ω) consisting of indecomposable preprojective objects, preinjective objects, and the first shifts of indecomposable projective modules to the set of cluster variables of the corresponding cluster algebra 𝒜_Ω is given. The images are called “preprojective cluster variables”. It is proved that all preprojective cluster variables other than u_i have denominators u ^dim M in their irreducible fractions of integral polynomials, where M is the corresponding preprojective module or preinjective module. In case the valued quiver (Γ, Ω) is of finite type, the denominator theorem holds with respect to any cluster. Namely, let x = (x₁,…, x_n) be a cluster of the cluster algebra 𝒜_Ω, and V the cluster tilting object in 𝒞(Ω) corresponding to x, whose endomorphism algebra is denoted by Λ. Then the denominator of any cluster variable y other than x_i is x ^dim M, where M is the indecomposable Λ-module corresponding to y. This result is a generalization of the corresponding result of Caldero–Chapoton–Schiffler to the non-simply-laced case.