一种量子自组织特征映射网络模型及聚类算法

提出一种量子自组织特征映射网络模型及聚类算法.量子神经元的输入和权值均为量子比特,输出为实数,量子自组织特征映射网络由输入层和竞争层组成.首先将聚类样本转换成量子态形式并提交给输入层,完成聚类样本的输入;然后计算样本量子态与相应权值量子态的相似系数,提取聚类样本所隐含的模式特征,并对其进行自组织,在竞争层将聚类结果表现出来.采用量子门更新量子权值,分无监督和有监督两个阶段完成网络的训练.仿真实验结果表明该模型及算法明显优于普通自组织特征映射网络.     

基于自组织模糊神经网络的污水处理出水水质预测建模

提出了一种新的自组织模糊神经网络算法,该算法能够基于输入数据自动进行神经网络结构辨识和参数辨识。首先采用一种自组织聚类方法得到神经网络的结构和网络参数初值,然后采用监督学习来优化网络参数。以某污水处理厂的运行数据为对象,应用该自组织模糊神经网络建立了活性污泥污水处理系统出水水质预测模型。仿真结果表明,该模型能够对污水处理系统出水水质进行较好的预测。     

基于聚类模糊神经网络的非线性电路故障诊断

提出了一种基于聚类算法和模糊神经网络的非线性模拟电路故障诊断方法。通过一个无监督的聚类算法自组织地确定模糊规则的数目并生成一个初始的故障诊断模糊规则库，构造了一类模糊神经网络，通过训练调整网络权值，使故障诊断模糊规则库的分类更加精确，并通过仿真实验验证了该方法的有效性。     

基于改进的自组织映射入侵检测方法

入侵检测作为保障互联网安全的主要措施之一,对于网络入侵的识别和诊断有着重要的意义。将自组织映射(SOM)的思想引入网络入侵检测中,提出了一种基于SOM的网络入侵检测算法。算法通过对SOM神经网络中输出神经元的邻域密度进行排名,同时结合受试者工作特征(ROC)曲线设置邻域密度阈值等方法,使得入侵检测的结果通过输出神经元的邻域密度进行表达,克服了SOM神经网络训练时容易产生畸变导致输出神经元自身的聚类结果难以理解的缺点。通过对算法仿真实验,表明该算法不仅有效而且拥有相当可观的检测率。     

SOM-LSTM递归神经网络语音端点检测系统

语音端点检测是语音信号处理的一个重要环节,传统的语音端点检测算法往往是基于短时能量以及过零率等实现,在低信噪比的环境下,检测算法的准确度较低。因此,提出了一种基于自组织映射(SOM)神经网络和长短时记忆(LSTM)递归神经网络相结合的端点检测算法。该算法通过检测语音信号在每个时间节点上的特征属性利用SOM神经网络进行聚类,并根据每个时间节点的语音状态对聚类结果进行调整,构造能够判别语音状态和噪声状态的样本作为LSTM递归神经网络的输入,利用LSTM递归神经网络实现端点检测的目的。     

基于自组织特征映射神经网络的聚类分析

在深入研究自组织特征映射（Self-organizing Feature Mapping,SOFM）神经网络的结构和聚类算法的基础上,阐述了SOFM网络的建立方法.以随机二维向量的聚类为例,利用所建立的SOFM网络模型对输入的随机二维向量进行聚类,并着重研究了输出层神经元拓扑结构、训练步数对聚类结果的影响以及在相同拓扑结构条件下,SOFM网络模型的权值向量的调整过程.仿真结果表明：在输出层神经元节点形式为六边型条件下,输出层神经元的个数越多,SOFM网络模型的聚类结果就越准确;在相同的拓扑结构条件下,训练步数越大,SOFM网络聚类结果越准确,但过大的训练步数对于聚类结果的影响甚微.     

基于优化RBF神经网络的无线室内定位

针对径向基函数（Radial Basis Function,RBF）神经网络算法在无线网络室内定位中拓扑结构和网络参数难以确定,其定位效果不理想的问题,提出了一种用核主成分分析的模糊C均值聚类算法（Fuzzy C-Means clustering algorithm based on Kernel Principal Component Analysis,KPCA-FCM）和模拟退火自适应遗传算法（Simulated Annealing adaptive Genetic Algorithm,SAGA）优化RBF神经网络的无线室内定位算法。首先利用KPCA对原始训练数据样本进行数据预处理,再通过KPCA-FCM算法计算出最优聚类数目和聚类中心点;其次将聚类数目和聚类中心点作为隐含层神经元个数和中心值,创建RBF神经网络,并将其网络参数映射到SAGA算法中;再次由SAGA算法进行网络参数寻优,把最优的解映射回RBF神经网络;最后利用样本数据对RBF神经网络进行训练和测试,完成建立RBF神经网络算法模型。实验表明,在相同的环境中,所提算法比传统RBF神经网络定位精度提高了48.41%。     

基于神经网络的几何模型区域分割算法

三维模型简化是近年来计算机图形学中的一个研究热点,现有的简化算法多从全局出发,对几何模型的各个部位统一进行简化,因此模型简化后大量的细节特征丢失.针对三维模型简化中保留细节特征的需要,提出了一种基于自组织特征映射神经网络的三维模型区域分割算法.首先计算三维几何模型中每一顶点的特征向量,然后利用该向量作为自组织特征映射神经网络的输入模式实现对三维模型的聚类分割,最后采取提出的相关性最大准则对过分割区域进行合并,得到最终分割结果.实验表明,该方法能有效地分割出模型的细节区域,满足三维模型简化中保留细节特征的需要.     

军用无人机研制费用的RBF神经网络预测

应用基于最近邻聚类算法的径向基函数(RBF)网络建立了军用无人机研制费用预测模型,并采用该模型对某型军用无人机研制费用进行了预测。实例表明,与多元线性回归和BP神经网络的预测结果相比,建立的新型军用无人机研制费用预测模型具有更高的预测精度。     

基于BP神经网络的RSSI测距优化算法

基于接收信号强度指示(received signal strength indication, RSSI)测距的研究和应用领域很广泛,一直是物联网研究的热点. 为降低传统基于反向传播(back propagation,BP)神经网络的RSSI测距误差,文中提出一种基于K-means聚类算法对样本数据进行预处理的BP神经网络测距算法,来解决由于RSSI值衰减程度不同引起的不同距离区间RSSI值和真实距离之间映射关系不均匀的问题. 将K-means聚类算法应用于BP神经网络模型中,对样本数据进行距离区间划分,然后将已经分类好的数据分别输入BP神经网络建立网络模型并进行实验仿真. 结果显示:传统基于BP神经网络的RSSI测距算法的均方根误差为1.425 7 m;而经过K-means算法改进后的BP神经网络测距算法的均方根误差为1.288 7 m,降低了测距误差,并优化了目标RSSI值与真实距离的映射关系.     

Clustering of DOA data in radar pulse based on SOFM and CDbw

Clustering is the main method of deinterleaving of radar pulse using multi-parameter. However, the problem in clustering of radar pulses lies in finding the right number of clusters. To solve this problem, a method is proposed based on Self-Organizing Feature Maps （SOFM） and Composed Density between and within clusters （CDbw）. This method firstly extracts the feature of Direction Of Arrival （DOA） data by SOFM using the characteristic of DOA parameter, and then cluster of SOFM. Through computing the cluster validity index CDbw, the right number of clusters is found. The results of simulation show that the method is effective in sorting the data of DOA.     

Unsupervised multistage image classification using hierarchical clustering with a Bayesian similarity measure.

A new multistage method using hierarchical clustering for unsupervised image classification is presented. In the first phase, the multistage method performs segmentation using a hierarchical clustering procedure which confines merging to spatially adjacent clusters and generates an image partition such that no union of any neighboring segments has homogeneous intensity values. In the second phase, the segments resulting from the first stage are classified into a small number of distinct states by a sequential merging operation. The region-merging procedure in the first phase makes use of spatial contextual information by characterizing the geophysical connectedness of a digital image structure with a Markov random field, while the second phase employs a context-free similarity measure in the clustering process. The segmentation procedure of region merging is implemented as a hierarchical clustering algorithm whereby a multiwindow approach using a pyramid-like structure is employed to increase computational efficiency while maintaining spatial connectivity in merging. From experiments with both simulated and remotely sensed data, the proposed method was determined to be quite effective for unsupervised analysis. In particular, the region-merging approach based on spatial contextual information was shown to provide more accurate classification of images with smooth spatial patterns.     

Analysis of Scattering in Mobile Radio Channels Based on Clustered Multipath Estimates

This paper presents the results of angle and delay measurements in physically nonstationary radio channels obtained in an outdoor urban environment. The multidimensional estimation data are obtained using a recently developed 3-D high-resolution channel sounder. The estimation results are compared with results obtained from a 3-D deterministic propagation prediction tool. For a better analysis, a hierarchical clustering method is presented that can separate and group the multidimensional estimation data into clusters. Measurements performed at a fixed position as well as along a trajectory are used to characterize the angular dispersion in both azimuth and elevation. The angular dispersion in terms of the rms cluster angular spread in both elevation and azimuth of the different clusters is analyzed over space and time and related to its physical scattering sources. Compared to the measurements, a large number of multipath clusters are missing in the predictions. Furthermore, it is observed from the measurements that different objects cause different angular spread values in azimuth and elevation. The results can be very helpful for the identification, improvement and calibration of deterministic propagation prediction models.     

电子地图色域映射的SOM神经网络方法

颜色的设备相关性导致电子地图在跨媒介再现时容易出现色彩变形以及地理信息错误传输等问题.本文分析了电子地图色域的特征,提出了一种基于SOM(Self-Organizing Maps)神经网络的色域映射方法.针对神经网络方法没有顾及电子地图色域映射过程中的色彩空间权重以及神经网络中邻域函数的各向同性假设不适用于地图色域这两个问题,提出了改进方法.将本文方法与ICC(International Color Consortium)感知再现、绝对色度再现色域映射方法进行了对比,结果表明本文方法能够很好的保持地图整体色差,能够提高电子地图颜色复制精度.     

基于密度估计的社会网络特征簇挖掘方法

通过凝聚式聚类方法抽取网络的层次结构,并基于拓扑结构分析,给出了社会网络的标注密度估计函数。通过对密度估计函数在网络层次结构上的聚合操作,计算聚簇的特征性指标,从而达到发现特征聚簇的目的。在大规模的真实数据上对这些方法和模型进行了验证,实验结果表明,所提出的思路和模型是合理的,算法是高效、可伸缩的。     

基于属性关系矩阵的AP子空间聚类算法

Affinity Propagation（AP）聚类算法将所有数据点作为潜在的聚类中心，在相似度矩阵的基础上通过消息传递进行聚类, 但却不适用于子空间聚类。基于属性关系矩阵的AP子空间聚类算法（AP clustering algorithm based on attributes relation matrix, ARMAP）是一种异步软子空间聚类算法，首先通过计算属性a的 邻域得到属性的关系矩阵，然后通过查找极大全1子矩阵得到数据集的兴趣度子空间，最后在各兴趣度子空间使用AP算法聚类，完成子空间聚类的任务。ARMAP算法将子空间的查找转换成查找矩阵的极大全1子矩阵，在正确查找子空间的同时，降低了时间复杂度。算法既保留了AP聚类算法的优点，又克服了AP算法不能进行子空间聚类的不足。     

数据流上基于K-median聚类的算法研究

文章研究和分析了数据流上的K-median聚类算法技术,包括:(1)流模型和K-median问题定义;(2)基于流的K-median聚类基本决策和内在机理;(3)理论上有性能保证的流算法。对于每一特征,这种技术能在没有实际保留任何数据流对象的情形下有效地确定聚类点。它通过一个聚类块的一分为二或相邻聚类块的合二为一来动态地生成聚类点,从而实现上述目标。作为结果,这种技术所确定的聚类点将比其他常规方法更准确。在数据流环境中,这种技术能够在产生高质量聚类结果的同时非常有效地执行。     

Fault clustering: modeling and observation on experimental LSI chips

Occurrence of fault clustering on large-scale integrated (LSI) MOS product was verified with optical microscopes on experimental chips that failed electrical testing. Two methods were used for determining clustering: analysis of the fault density derived from collected fault data, and separation of faults into two populations, one representing solitary faults, the other clusters. A model for the first method is presented and its effectiveness examined on a simulated fault set. The method is then applied to fault data representing two samples of MOS LSI experimental product. Population separation is finally carried out on one of the data samples, and the clustering data developed from this process are expressed by two factors. One factor can be used for refined yield estimates, the other was applied to quality measure calculations.     

A Novel Approach for Discovering Overlapping Clusters in Gene Expression Data

Many existing clustering algorithms have been used to identify coexpressed genes in gene expression data. These algorithms are used mainly to partition data in the sense that each gene is allowed to belong only to one cluster. Since proteins typically interact with different groups of proteins in order to serve different biological roles, the genes that produce these proteins are therefore expected to coexpress with more than one group of genes. In other words, some genes are expected to belong to more than one cluster. This poses a challenge to gene expression data clustering as there is a need for overlapping clusters to be discovered in a noisy environment. For this task, we propose an effective information theoretical approach, which consists of an initial clustering phase and a second reclustering phase, in this paper. The proposed approach has been tested with both simulated and real expression data. Experimental results show that it can improve the performances of existing clustering algorithms and is able to effectively uncover interesting patterns in noisy gene expression data so that, based on these patterns, overlapping clusters can be discovered.     

A KNN-Scoring Based Core-Growing Approach to Cluster Analysis

This paper proposes a novel core-growing (CG) clustering method based on scoring k-nearest neighbors (CG-KNN). First, an initial core for each cluster is obtained, and then a tree-like structure is constructed by sequentially absorbing data points into the existing cores according to the KNN linkage score. The CG-KNN can deal with arbitrary cluster shapes via the KNN linkage strategy. On the other hand, it allows the membership of a previously assigned training pattern to be changed to a more suitable cluster. This is supposed to enhance the robustness. Experimental results on four UCI real data benchmarks and Leukemia data sets indicate that the proposed CG-KNN algorithm outperforms several popular clustering algorithms, such as Fuzzy C-means (FCM) (Xu and Wunsch IEEE Transactions on Neural Networks 16:645–678, 2005), Hierarchical Clustering (HC) (Xu and Wunsch IEEE Transactions on Neural Networks 16:645–678, 2005), Self-Organizing Maps (SOM) (Golub et al. Science 286:531–537, 1999; Tamayo et al. Proceedings of the National Academy of Science USA 96:2907, 1999), and Non-Euclidean Norm FCM (NEFCM) (Karayiannis and Randolph-Gips IEEE Transactions On Neural Networks 16, 2005).