一种新的基于图论聚类的分割算法

针对传统图论聚类法在分割图像时对噪声和模糊边界敏感,产生伪割集以及计算复杂度大的问题,对传统算法进行了相应的改进,即首先将每个像素作为一类改为将图像中灰度相同的像素作为一类;其次在计算权值时改进权函数定义,将节点与区域间的空间近邻关系约束进权函数表达式,而非传统算法中仅考虑节点与节点间的灰度和位置关系.对比实验表明,该算法只需要设计少量的参数即可自动完成聚类,所需的存储空间以及实现的复杂度相比于传统图论聚类法都得到极大改善.     

谱聚类中的相似度矩阵研究

以谱聚类算法为理论基础,对图的分割问题构造合适的相似度矩阵。构造四种相似度矩阵,并进行实验分析。实验结果表明,好的相似度矩阵对图的分割是非常有效的。     

基于模式聚类的数据预处理方法应用研究

在分析数据预处理的意义基础上，提出了一种基于最大距离算法的模式聚类的数据预处理方法。该方法不依赖于任何数学模型，通过对某造纸厂大量数据的仿真处理，研究表明本文提出的方法能在保留原始数据的有用信息的基础上剔除冗余数据，侦破过失误差，减少随机误差。     

基于MapReduce的图结构聚类算法

图结构聚类（SCAN）是一种著名的基于密度的图聚类算法。该算法不仅能够找到图中的聚类结构,而且还能发现图中的Hub节点和离群节点。然而,随着图数据规模越来越大,传统的SCAN算法的复杂度为O（m^1.5）（m为图中边的条数）,因此很难处理大规模的图数据。为了解决SCAN算法的可扩展性问题,本文提出了一种新颖的基于MapReduce的海量图结构聚类算法MRSCAN。具体地,我们提出了一种计算核心节点,以及两种合并聚类的MapReduce算法。最后,在多个真实的大规模图数据集上进行实验测试,实验结果验证了算法的准确性、有效性,以及可扩展性。     

Unsupervised and simultaneous training of multiple object detectors from unlabeled surveillance video

Object detection is an essential component in automated vision-based surveillance systems. In general, object detectors are constructed using training examples obtained from large annotated data sets. The inevitable limitations of typical training data sets make such supervised methods unsuitable for building generic surveillance systems applicable to a wide variety of scenes and camera setups. In our previous work we proposed an unsupervised method for learning and detecting the dominant object class in a general dynamic scene observed by a static camera. In this paper, we investigate the possibilities to expand the applicability of this method to the problem of multiple dominant object classes. We propose an idea on how to approach this expansion, and perform an evaluation of this idea using two representative surveillance video sequences.     

基于环型网络模体应用马尔科夫聚类的图挖掘模型*

针对图数据挖掘效率低、精度低等问题,提出基于环型网络模体应用马尔科夫聚类的图挖掘模型.首先,依据输入图的点集,采用Erdo″s-Rényi模型生成随机图,在输入图和随机图的子图挖掘过程中利用向量的加法性质判定环型子图,计算网络模体的统计特征,判断子图是否为网络模体.然后,求解图中边的绝对贡献值关联矩阵,通过动态阈值法求得阈值,二值化处理该矩阵.最后,对已稀疏化的图进行扩张和膨胀操作,使其达到收敛状态.实验表明,文中模型有效减少运行时间,在保证聚类质量同时提高图挖掘效率.     

基于图切割和密度聚类的视频行人检测算法*

现有视频行人检测方法把行人检测看成一个有监督的两类(即行人和背景)学习问题,区分视频中的行人和背景,并不能很好解决行人的姿态变化和行人间的遮挡问题.文中提出基于图切割和密度聚类的行人检测算法,把行人检测看成一个多类的无监督学习过程.在训练阶段,首先对每个训练样本计算多级梯度方向直方图-局部二分模式(HOG-LBP)特征,然后对多级HOG-LBP特征所属的每个图像块分配不同的权值.为了区别行人的不同部位并赋权值,采用基于图像块的图分割方法从背景中分割行人所在的图像块.最后,再采用基于密度峰值的聚类算法对正样本和负样本分别进行无监督的聚类.在测试阶段,首先通过计算样本特征与每个聚类中心的距离,然后使用前5个最短距离进行投票,判断其是否包含行人.实验证明,文中算法较好解决行人的姿态变化和行人间的遮挡问题,并且随着训练样本的增加,能取得和目前最优行人检测方法可比较的结果.     

一种新的基于图划分聚类算法---GAGPBCUK算法

提出了一种新的基于图划分的聚类算法——GAGPBCUK算法。该算法解决了谱聚类算法参数敏感和聚类结果不准确等问题。3组仿真实验结果表明,GAGPBCUK算法不仅在识别和学习数据集中的隐含聚类数方面具有很好的性能,而且能够得到比谱聚类算法(NJW算法)更加有效的聚类结果。     

GA-based multiple paths test data generator

Developers have learned over time that software testing costs a considerable amount of a software project budget. Hence, software quality managers have been looking for solutions to reduce testing costs and time. Considering path coverage as the test adequacy criterion, we propose using genetic algorithms (GA) for automating the generation of test data for white-box testing. There are evidences that GA has been already successful in generating test data. However, existing GA-based test data generators suffer from some problems. This paper presents our approach to overcome one of these problems; that is the inefficiency in covering multiple target paths. We have designed a GA-based test data generator that is, in one run, able to synthesize multiple test data to cover multiple target paths. Moreover, we have implemented a set of variations of the generator. Experimental results show that our test data generator is more efficient and more effective than others.     

Chordless paths through three vertices

Consider the following problem, which we call “Chordless path through three vertices” or CP3V, for short: Given a simple undirected graph G=(V,E)$G=(V,E)$, a positive integer k, and three distinct vertices s, t  , and v∈V$v\in V$, is there a chordless path of length at most k from s   via v$v$ to t in G? In a chordless path, no two vertices are connected by an edge that is not in the path. Alternatively, one could say that the subgraph induced by the vertex set of the path in G is the path itself. The problem has arisen in the context of service deployment in communication networks. We resolve the parametric complexity of CP3V by proving it W[1]$W[1]$-complete with respect to its natural parameter k  . Our reduction extends to a number of related problems about chordless paths and cycles. In particular, deciding on the existence of a single directed chordless (s,t)$(s,t)$-path in a digraph is also W[1]$W[1]$-complete with respect to the length of the path.     

一种基于增长模式的交易序列聚类算法

对交易序列进行各种挖掘分析能为商家制定营销策略提供量化依据.文中从销售量及变化趋势角度研究交易序列数据集的内在结构,定义了一种反映价格变化趋势的增长模式及其错位组合距离和角度向量距离两种相似性度量,在此基础上设计一个考虑时限约束的目标函数进行聚类研究.实验数据采用真实的商品交易序列集,结果表明,在时限约束的条件下,增长模式这种特征提取方式及其模式间的两种距离函数能较好地产生聚类结果,且这些聚类结果能得到较好地解释.     

基于数据流的聚类趋势分析算法

聚类趋势分析算法基于抽样原理导致聚类趋势指标不稳定和片面,而且不适应数据流的批量增量特性,因而需要重复进行聚类趋势指数计算。为此,基于全体数据进行整体分析,提出一种基于最小距离连通图（MDCG）的聚类趋势分析算法MDCG-CTI。首先,利用栈的深度优先遍历法更新增量数据的最邻近路径从而降低MDCG的建立复杂度;然后,计算聚类趋势指数并确定可聚类性的判定阈值;最后,将所提算法和批量增量的具有噪声的基于密度的聚类方法（DBSCAN）相结合。在自定义数据集上的实验表明,该算法比现有算法对单簇和含大量噪点的数据的可聚类性判断更为精确;而在大数据集pendigits和avila上,所提算法比基于谱方法的聚类趋势可视化分析（SpecVAT）累计耗时降低了38%和42%,且相较SpecVAT结合批量增量DBSCAN,该算法结合批量增量DBSCAN的聚类平均准确率分别提高了6%和11%,聚类累计耗时则分别降低了7%和8%。实验结果表明该算法可以准确无参地判断聚类趋势,并明显提高增量聚类的有效性和运行效率。     

Finding multiple induced disjoint paths in general graphs

In an undirected graph, paths P₁,P₂,…,Pk are induced disjoint if each one of them is chordless (i.e., is an induced path) and any two of them have neither common nodes nor adjacent nodes. This paper investigates the Maximum Induced Disjoint Paths (MIDP) problem: in an undirected graph G=(V,E), given k node pairs {s₁,t₁},…,{sk,tk}, connect maximum number of these node pairs via induced disjoint paths. Till now, the only things known about MIDP are: i) it is NP-hard; ii) it is NP-hard even when k=2; iii) it can be solved in polynomial time when k is a fixed constant and the given graph is a directed planar graph (Kobayashi, 2009 [9]). This paper proves that for general k and any ?>0, it is NP-hard to approximate MIDP within m1/2−?, where m=|E|. Two algorithms for MIDP are given by this paper: a greedy algorithm whose approximation ratio is and an on-line algorithm which has a good lower bound.     

基于图收缩的半监督聚类算法

为了在只有少量已知标记的数据集中获得较好的聚类效果,提出了一种基于图收缩的半监督聚类算法。首先将整个样本空间中的数据表达为一个带权图,再根据给出的must-link约束,对图进行边收缩的修改,进而增强must-link约束。在此基础上引入图拉普拉斯算子,结合cannot-link约束将样本空间投影到一个特征子空间。最后在子空间上进行聚类分析。实验结果表明,该方法不仅提高了对复杂数据的聚类结果,而且在约束对数量较少时也能获得较好的结果。     

Clustering of wireless sensor and actor networks based on sensor distribution and connectivity

Wireless Sensor and Actor Networks (WSANs) employ significantly more capable actor nodes that can collect data from sensors and perform application specific actions. To take these actions collaboratively at any spot in the monitored regions, maximal actor coverage along with inter-actor connectivity is desirable. In this paper, we propose a distributed actor positioning and clustering algorithm which employs actors as cluster-heads and places them in such a way that the coverage of actors is maximized and the data gathering and acting times are minimized. Such placement of actors is done by determining the k-hop Independent Dominating Set (IDS) of the underlying sensor network. Basically, before the actors are placed, the sensors pick the cluster-heads based on IDS. The actors are then placed at the locations of such cluster-heads. We further derive conditions to guarantee inter-actor connectivity after the clustering is performed. If inter-connectivity does not exist, the actors coordinate through the underlying sensors in their clusters to adjust their locations so that connectivity can be established. The performances of the proposed approaches are validated through simulations.     

基于协同聚类的两阶段文本聚类方法

为了将语义信息用于文本聚类和有效地进行特征选择,文中提出一种基于协同聚类的两阶段文本聚类方法.该方法分别对文档和特征进行聚类从而得到特征与主题之间的语义关联关系.然后利用此关系来相互调整彼此的聚类结果.实验结果表明,利用特征与主题之间的语义关联关系能有效提高聚类效果.     

谱聚类算法综述

谱聚类算法是近年来国际上机器学习领域的一个新的研究热点.谱聚类算法建立在谱图理论基础上,与传统的聚类算法相比,它具有能在任意形状的样本空间上聚类且收敛于全局最优解的优点.本文首先介绍了图论方法用于聚类的基本理论,然后根据图划分准则对谱聚类算法进行分类,着重阐述了各类中的典型算法,并对算法进行了比较分析,最后进行总结并提出了几个有价值的研究方向.     

基于模糊聚类算法的神经网络集成

基于模糊聚类思想,提出了一种神经网络集成方法。利用隶属度函数,构造了一个分布函数,根据分布函数对训练数据进行抽样,用所抽得的数据作为个体神经网络的训练样本,多个个体神经网络构成神经网络集成,集成的输出采用相对多数投票法。理论分析和实验结果表明,该方法对模式分类能取得较好的效果。     

基于一类分类的聚类方法及其应用

在分析一类分类方法最新研究成果的基础上,提出了基于一类分类的模式聚类方法,通过与传统聚类方法的比较,说明该方法对非线性数据处理聚类的优越性,并以某企业对供应商关系调查数据为例,将这种方法应用于企业供应商关系分析中,聚类结果表明了该方法的有效性,为企业与其它组织之间的复杂关系网络的特征数据分析提供了一种可行的方法。     

Assessment of Microarray Data Clustering Results Based on a New Geometrical Index for Cluster Validity

A measurement of cluster quality is often needed for DNA microarray data analysis. In this paper, we introduce a new cluster validity index, which measures geometrical features of the data. The essential concept of this index is to evaluate the ratio between the squared total length of the data eigen-axes with respect to the between-cluster separation. We show that this cluster validity index works well for data that contain clusters closely distributed or with different sizes. We verify the method using three simulated data sets, two real world data sets and two microarray data sets. The experiment results show that the proposed index is superior to five other cluster validity indices, including partition coefficients (PC), General silhouette index (GS), Dunn’s index (DI), CH Index and I-Index. Also, we have given a theorem to show for what situations the proposed index works well.