图优化的低秩双随机分解聚类

低秩双随机矩阵分解聚类(low-rank doubly stochastic matrix decomposition for cluster analysis，DCD)是最近由Yang等人[16]提出的一种图聚类方法，它通过最小化KL(Kullback-Leibler)散度准则：KL(A，S)，从图关联矩阵S中获得一个非负低秩双随机矩阵分解：A=UUT(U(0)，并以U作为类标签矩阵进行聚类。在DCD方法中，因矩阵S是固定不可变的，故S初始取值选取的好坏对聚类结果有极大影响，这导致了它缺乏稳定性。针对这一问题，提出了一种基于图优化的DCD方法，将图关联矩阵S和DCD的优化集成在统一框架中，这改进和拓展了原始的DCD方法。实验结果表明，与DCD方法相比，图优化的DCD方法具有更好的聚类精确度和稳定性。     

Correlation-based spectral clustering for flexible process monitoring

The individuality of production devices should be taken into account when statistical models are designed for parallelized devices. In the present work, a new clustering method, referred to as NC-spectral clustering, is proposed for discriminating the individuality of production devices. The key idea is to classify samples according to the differences of the correlation among measured variables, since the individuality of production devices is expressed by the correlation. In the proposed NC-spectral clustering, the nearest correlation (NC) method and spectral clustering are integrated. The NC method generates the weighted graph that expresses the correlation-based similarities between samples, and the constructed graph is partitioned by spectral clustering. A new statistical process monitoring method and a new soft-sensor design method are proposed on the basis of NC-spectral clustering. The usefulness of the proposed methods is demonstrated through a numerical example and a case study of parallelized batch processes.     

基于自编码器的贝叶斯网嵌入及概率推理

贝叶斯网(BN)是不确定性知识表示和推理的基本框架,广泛用于社交网络、知识图谱和医疗诊断等领域.特定领域中基于BN的分析诊断和决策支持,其核心计算任务是基于BN进行多次概率推理.然而,使用传统的概率推理方法,基于同一BN的多次概率推理其中间过程存在很多重复的计算结果,具有较高的时间复杂度.为了提高多次概率推理的效率,提出易于重用和易于计算的贝叶斯网嵌入及相应的概率推理方法.首先,借鉴图嵌入的基本思想,使用点互信息矩阵来表示BN的有向无环图结构和条件概率参数,提出基于自编码器和注意力机制的BN嵌入方法.其中,自编码器的每一编码层利用节点与其邻居节点(父节点和子节点)的相关性生成节点嵌入,从而在嵌入向量中保存BN节点间的概率依赖关系.然后,使用嵌入向量之间的距离来度量节点之间的联合概率,提出基于嵌入向量的BN概率推理方法.实验证明,针对BN的多次概率推理,所提方法的效率高于现有方法,且能得到准确的推理结果.     

专家证据文档识别无向图模型

专家证据文档识别是专家检索的关键步骤.融合专家候选文档独立页面特征以及页面之间的关联关系,提出了一个专家证据文档识别无向图模型.该方法首先分析各类专家证据文档中的词、URL 链接、专家元数据等独立页面特征以及候选专家证据文档间的链接和内容等关联关系;然后将独立页面特征以及页面之间的关联关系融入到无向图中构建专家证据文档识别无向图模型;最后利用梯度下降方法学习模型中特征的权重,并利用吉布斯采样方法进行专家证据文档识别.通过对比实验验证了该方法的有效性.实验结果表明,该方法有较好的效果.     

SINBAD automation of scientific discovery: From factor analysis to theory synthesis

Modern science is turning to progressively more complex and data-rich subjects, which challenges the existing methods of data analysis and interpretation. Consequently, there is a pressing need for development of ever more powerful methods of extracting order from complex data and for automation of all steps of the scientific process. Virtual Scientist is a set of computational procedures that automate the method of inductive inference to derive a theory from observational data dominated by nonlinear regularities. The procedures utilize SINBAD – a novel computational method of nonlinear factor analysis that is based on the principle of maximization of mutual information among non-overlapping sources, yielding higher-order features of the data that reveal hidden causal factors controlling the observed phenomena. The procedures build a theory of the studied subject by finding inferentially useful hidden factors, learning interdependencies among its variables, reconstructing its functional organization, and describing it by a concise graph of inferential relations among its variables. The graph is a quantitative model of the studied subject, capable of performing elaborate deductive inferences and explaining behaviors of the observed variables by behaviors of other such variables and discovered hidden factors. The set of Virtual Scientist procedures is a powerful analytical and theory-building tool designed to be used in research of complex scientific problems characterized by multivariate and nonlinear relations. This revised version was published online in June 2006 with corrections to the Cover Date.     

基于二分图的RDF关键词扩展查询方法

使用图表示RDF数据可以保持数据间的关联信息和语义信息,越来越多的关键词查询方法基于图结构实现RDF数据的查询处理。将二分图与RDF数据图相结合,定义RDF二分图模型,并提出一种基于二分图的RDF关键词扩展查询方法KERBG。该方法将文本信息封装在二分图顶点标签上,以支持对关系的查询;利用关键词同义词扩展技术对查询关键词进行语义扩展,有效解决同一对象的描述用词的多样性问题,进而提高查准率;利用RDF二分图的反对称邻接矩阵及其幂矩阵构造包含关键顶点的查询结果子图,实现关键词查询处理,并降低查询响应时间。实验结果表明,在查准率和查询响应时间方面,提出的KERBG方法优于当前主流方法。     

基于深度图正则化矩阵分解的多视图聚类算法

针对现实社会中由多种表示或视图组成的多视图数据广泛存在的问题,深度矩阵分解模型因其能够挖掘数据的层次信息而备受关注,但该模型忽略了数据的几何结构信息.为解决以上问题,本文提出基于深度图正则化矩阵分解的多视图聚类算法,通过获取每个视图的局部结构信息和全局结构信息在逐层分解中加入两个图正则化限制,保护多视图数据的几何结构信...     

Cross grouping strategy based 2DPCA method for face recognition

Grouping strategy exactly specifies the form of covariance matrix, therefore it is very essential. Most 2DPCA methods use the original 2D image matrices to form the covariance matrix which actually means that the strategy is to group the random variables by row or column of the input image. Because of their grouping strategies these methods have two main drawbacks. Firstly, 2DPCA and some of its variants such as A2DPCA, DiaPCA and MatPCA preserve only the covariance information between the elements of these groups. This directly implies that 2DPCA and these variants eliminate some covariance information while PCA preserves such information that can be useful for recognition. Secondly, all the existing methods suffer from the relatively high intra-group correlation, since the random variables in a row, column, or a block are closely located and highly correlated. To overcome such drawbacks we propose a novel grouping strategy named cross grouping strategy. The algorithm focuses on reducing the redundancy among the row and the column vectors of the image matrix. While doing this the algorithm completely preserves the covariance information of PCA between local geometric structures in the image matrix which is partially maintained in 2DPCA and its variants. And also in the proposed study intra-group correlation is weak according to the 2DPCA and its variants because the random variables spread over the whole face image. These make the proposed algorithm superior to 2DPCA and its variants. In order to achieve this, image cross-covariance matrix is calculated from the summation of the outer products of the column and the row vectors of all images. The singular value decomposition (SVD) is then applied to the image cross-covariance matrix. The right and the left singular vectors of SVD of the image cross-covariance matrix are used as the optimal projective vectors. Further in order to reduce the dimension LDA is applied on the feature space of the proposed method that is proposed method + LDA. The exhaustive experimental results demonstrate that proposed grouping strategy for 2DPCA is superior to 2DPCA, its specified variants and PCA, and proposed method outperforms bi-directional PCA + LDA.     

一种潜在信息约束的非负矩阵分解方法

传统的非负矩阵分解方法没有充分利用数据间的内在相似性,从而影响了算法的性能。为此,本文提出一种潜在信息约束的非负矩阵分解方法。该方法首先利用迭代最近邻方法挖掘原始数据的潜在信息,然后利用潜在信息构造数据之间的相似图,最后将相似图作为约束项求得非负矩阵的最优分解。相似图的约束使得非负矩阵分解在降维过程中保持了原始数据之间的相似性关系,进而提高了非负矩阵分解的判别能力。图像聚类实验结果表明了该方法的有效性。     

Root sensitivity to parameter uncertainties: a statistical approach

A statistical approach to calculating the sensitivity of characteristic roots to plant parameter uncertainties is presented. Means and variances of the eigenvalues are obtained in terms of the statistical moments of the system parameters, the sensitivity of an eigenvalue being defined by its relative standard deviation. The method allows for correlation among the system parameters and therefore among the elements of the plant matrix. The approach is a first-order random perturbation analysis that implicitly assumes gaussian random variables. The results derived by this statistical procedure are shown to be a generalization of the norm sensitivity results previously obtained in the literature by calculating the norm of the sensitivity matrix. Indeed, the two approaches are shown to be equivalent for uncorrelated and identically distributed plant matrix elements. The statistical measure of root sensitivity presented here is believed to be more representative than the usual norm sensitivity measure.     

基于LPCA的谱聚类算法

针对传统谱聚类在构建关系矩阵时只考虑样本的全局特征而忽略样本的局部特征、在聚类划分时通常需要指定聚类个数、无法对交叉点进行正确划分等问题,提出了一种改进的基于局部主成分分析和连通图分解的谱聚类算法。首先自动学习挑选数据集的中心点,然后使用局部主成分分析得到数据集的关系矩阵,最后用连通图分解算法完成对关系矩阵的划分。实验结果表明提出的改进算法性能优于现有经典算法。     

面向主动配电网实时优化调度的图强化学习方法

主动配电网的新能源、储能等能源形式可以有效提高运行的灵活性和可靠性, 同时新能源和负荷也给配电网带来了双重不确定性, 致使主动配电网的实时优化调度决策维度大、建模精度差. 针对这一问题, 本文提出结合图神经网络和强化学习的图强化学习方法, 避免对复杂系统的精准建模. 首先, 将实时优化调度问题表述为马尔可夫决策过程, 并将其表述为动态序贯决策问题. 其次, 提出了基于物理连接关系的图表示方法, 用以表达状态量的隐含相关性. 随后, 提出图强化学习来学习将系统状态图映射到决策输出的最优策略. 最后, 将图强化学习推广到分布式图强化学习. 算例结果表明, 图强化学习在最优性和效率方面都取得了更好的效果.     

多变量预测控制结构分解的图论方法

预测控制算法的计算复杂度主要由变量个数和控制时域决定, 而大型复杂系统中变量个数较多将导致计 算量大的问题, 尤其在有约束预测控制的优化求解中增加较重的计算负担. 本文针对此问题利用邻接矩阵、可达矩 阵和关联矩阵梳理系统传递函数模型中变量之间的关联, 将有关联的控制变量划分为一个子系统, 进而将一个大系 统分解成若干独立子系统, 即可将一个高维度的优化求解问题分解成多个维度较低的子优化问题, 降低计算复杂度 以达到减少计算量的目的. 最后将其应用在多变量有约束的双层结构预测控制算法中, 通过仿真进行验证.     

基于二部图的控制系统故障诊断方法

针对连续控制系统,建立了由系统约束集、变量集和边集构成的二部图模型,提出了一种定性描述与定量分析相结合的故障诊断算法。该算法通过分离子系统,求解系统的关联矩阵及最大匹配,定义了描述变量与系统约束之间依赖关系的规则,并设计了关联矩阵分层算法,以此来计算控制系统残差。以一个线性系统为例,探讨了该算法的应用过程,并通过仿真实例验证了该算法的有效性。     

融合共现推理的Faster R-CNN输电线路金具检测

为促进目标检测模型与电力领域业务知识有机融合,缓解金具样本间长尾分布问题,有效提升输电线路金具的自动化检测效果,提出了融合共现推理的Faster R-CNN(faster region-based convolutional neural network)输电线路金具检测模型。首先,深入研究输电线路金具结构化组合规则,通过数据驱动的方式以条件概率对金具目标间的共现连接关系进行有效表达;然后,结合图学习方法,利用学习并映射的共现概率关联作为共现图邻接矩阵,使用基线模型(Faster R-CNN)提取的特征向量作为图推理输入特征,以自学习的变换矩阵作为共现图传播权重,完成图信息传播并实现有效的共现推理模型嵌入。实验证明,融合共现推理模块的Faster R-CNN模型较原始模型提升了6.56%的准确率,对于长尾分布样本中数量较少的金具性能提升尤其显著。     

Scalability of correlation clustering

The problem of scalability of correlation clustering (CC) is addressed by reducing the number of variables involved in the SDP formulation. A nonlinear programming formulation is obtained instead of SDP formulation, which reduces the number of variables. The new formulation is solved through limited memory Broyden Fletcher Goldfarb Shanno method. We demonstrate the potential of the nonlinear formulation on large graph datasets having more than ten thousand vertices and nine million edges. The proposed scalable formulation is experimentally shown not to compromise on quality of the obtained clusters. We compare the scalable formulation results with those of the original CC formulation. We compare the scalable formulation results with the original CC formulation, a constrained spectral clustering method which uses edge labels of the graph and differs only in the way clusters are obtained by defining the cut on the given graph and with a variant of constraint spectral clustering known as self-taught spectral clustering.     

一种新颖的深度因果图建模及其故障诊断方法

为了实现复杂工业过程故障检测和诊断一体化建模, 提出了一种新颖的深度因果图建模方法. 首先, 利用循环神经网络建立深度因果图模型, 将Group Lasso稀疏惩罚项引入到模型训练中, 自动地检测过程变量间的因果关系. 其次, 利用模型学习到的条件概率预测模型对每个变量建立监测指标, 并融合得到综合指标进行整体工业过程故障检测. 一旦检测到故障, 对故障样本构建变量贡献度指标, 隔离故障相关变量, 并通过深度因果图模型的局部因果有向图诊断故障根源, 辨识故障传播路径. 最后, 通过田纳西?伊斯曼过程进行仿真验证, 实验结果验证了所提方法的有效性.     

A New Consensus Algorithm for Multi-Agent Systems via Decentralized Dynamic Output Feedback

In this paper, we study a consensus problem for multi-agent systems via dynamic output feedback control. The entire system is decentralized in the sense that each agent can only obtain output information from its neighbor agents. For practical purpose, we assume that actuator limitation exists, and require that the consensus be achieved among the agents at a specified convergence rate. By using an appropriate coordinate transformation, we reduce the consensus problem on hand to solving a strict matrix inequality, and then propose to use the homotopy based method for solving the matrix inequality. It turns out that our algorithm includes the existing graph Laplacian based algorithm as a special case.     

Information flow and cooperative control of vehicle formations

We consider the problem of cooperation among a collection of vehicles performing a shared task using intervehicle communication to coordinate their actions. Tools from algebraic graph theory prove useful in modeling the communication network and relating its topology to formation stability. We prove a Nyquist criterion that uses the eigenvalues of the graph Laplacian matrix to determine the effect of the communication topology on formation stability. We also propose a method for decentralized information exchange between vehicles. This approach realizes a dynamical system that supplies each vehicle with a common reference to be used for cooperative motion. We prove a separation principle that decomposes formation stability into two components: Stability of this is achieved information flow for the given graph and stability of an individual vehicle for the given controller. The information flow can thus be rendered highly robust to changes in the graph, enabling tight formation control despite limitations in intervehicle communication capability.     

Analogy-X: Providing Statistical Inference to Analogy-Based Software Cost Estimation

Abstract Data-intensive analogy has been proposed as a means of software cost estimation as an alternative to other data intensive methods such as linear regression. Unfortunately, there are drawbacks to the method. There is no mechanism to assess its appropriateness for a specific dataset. In addition, heuristic algorithms are necessary to select the best set of variables and identify abnormal project cases. We introduce a solution to these problems based upon the use of the Mantel correlation randomization test called Analogy-X. We use the strength of correlation between the distance matrix of project features and the distance matrix of known effort values of the dataset. The method is demonstrated using the Desharnais dataset and two random datasets, showing (1) the use of Mantel's correlation to identify whether analogy is appropriate, (2) a stepwise procedure for feature selection, as well as (3) the use of a leverage statistic for sensitivity analysis that detects abnormal data points. Analogy-X, thus, provides a sound statistical basis for analogy, removes the need for heuristic search and greatly improves its algorithmic performance.