Controlling the sparsity of a classifier is a key to train SVM efficiently on very large scale problems. This paper explores building SVM classifier on the fitting-plane of each class of data, which captures the distributing trend of the corresponding class of data. The newly developed plane-fitting model can be solved by core set methods, and the SVM is trained only on the core sets which are small subsets of the original data. The computing complexity of the proposed algorithm is up bounded by Ο(1/ε). Experimental results show that the new algorithm scales better than SVMperf and CVM/BVM, while their predicting accuracies are almost comparable.

The small number of support vectors is an important factor for SVM to fast deal with very large scale problems. This paper considers fitting each class of data with a plane by a new model, which captures separability information between classes and can be solved by fast core set methods. Then training on the core sets of the fitting-planes yields a very sparse SVM classifier. The computing complexity of the proposed algorithm is up bounded by \( {\text{\rm O}}(1/\varepsilon ) \). Experimental results show that the new algorithm trains faster than both CVM and SVMperf averagely, and with comparable generalization performance.

Abstract:In order to resolve the improper calculation method, an improvement on calculating interestingness was proposed by considering classified situations.Multiple templates were induced, which support multi-dimension analysis and enrich user’s implications. With two groups of contrast experiments, the new algorithm produced more reasonable sorting result of interestingness, and supplied users with more valuable rules.

It is practical and crucial that a designer will want to control the positions of some modules along the chip boundary in the final packing for I/O connection. To solve the problem we propose an algorithm named Area Estimate Transitive Closure Graphs (AE-TCG). By analyzing the feasible condition of boundary constraints, AE-TCG guarantees that the result of each perturbation is a feasible placement with boundary constraints, and doesn’t need to transform the infeasible solution to feasible one. Unlike most of the previous algorithms getting the target area after packing, AE-TCG can satisfy the boundary constraints and estimate the area of feasible placement without packing after random perturbation, then accept the beneficial perturbation. For the property of concentrate itself, AE-TCG is running without Simulated Annealing (SA) process. The experimental results show that AE-TCG is effective and efficient than other algorithms with boundary constraints in commonly used MCNC benchmark circuits.

Existed Affinity Propagation(AP) clustering algorithm can not reflect the clustering structure of the complex protein sequences, This paper proposes an adaptive AP classification method based on Generalized SMS and Huffman Decision(adAP/GSHD). Protein sequences are clustered via generalized Substitution Matching Similarity(gSMS) and existed adaptive affinity propagation(adAP) algorithm. It uses Huffman coding and confines the average code length of clustering results to embody the family clustering structure of protein sequences. By experiment of test adAP/GSHD and comparing its performance with other four classic clustering methods on six datasets of Clusters of Orthologous Groups(COG) of proteins database and Structural Classification of Proteins(SCOP) database, results demonstrate that this method not only can acquire number of clusters more approximately to the correct family number of clusters and more compact clustering structure for a given set of proteins, but also the average F-measure is 19.67%, 8.7%, 9.5%and 43.81%better than that of adAP, SMS, Spectral Clustering and TribeMCL respectively.

相对于硬聚类算法,软聚类算法可以更好地表示具有不精确边界的类簇。粗糙集和模糊集均是用于描述不确定数据的有效的数学工具,二者互为补充。研究人员已经将粗糙集和模糊集的概念相结合,并应用到聚类算法中,提出了粗糙模糊可能性C均值聚类算法。而文中通过引入阴影集,有效地解决了粗糙模糊可能性C均值聚类算法中的阈值选择问题。

A unsupervised feature annotation criterion-information gain criterion (IGC)-based on feature matrix information gain is proposed to rank the feature variable. According to this rank, three new feature filtering methods:direct selection (DS), cumulate maximum entropy (CEM), and information gain maximum (IGM) are given to reduce clustering complexity. The clustering results of these three filtering methods with two existing variance selection (VS) and gene shaving (GS) methods were tested and compared by using classic QC or K-means algorithm and three biological datasets: rod-shaped viruses (RSV), mixed-lineage leukemia (MLL), and acute leukemia patients (ALP). The experiment results show our feature filtering method has obvious superiority in accelerating the clustering procedure and preserving the clustering structure of initial data.

A data mining method MMHC was given to solve DNA sequences motifs.The seed-based mismatch clustering was used to form the candidate motif clusters.Then the depth first determination(DFD) algorithm based on relative entropy and cluster complexity was proposed to identify the true motif clusters.Finally,the conservation region scanning(CRS) and MAP value-preservation filtering(MVPF) were given to optimize motif clusters.The experiment was conducted by testing MMHC method and comparing its performance with other three classic motif discovery methods MEME,AlignACE and SOMBRERO on two classes of DNA sequences datasets.Experimental results show the superiority of MMHC method over the three classic motif discovery methods in reliability,precision and the reflection of the cluster structure of the background species for most of the DNA sequences datasets.

性能问题是阻碍DAA推广和应用的首要问题.为了进一步优化该协议的性能,找出性能瓶颈,定量地分析和测量DAA中各个实体的性能负荷分布是一个十分重要且必须的工作.文中详细分析了DAA的协议流程,提出了以机器周期为基本性能单位的性能负荷分布测量方法——归一化统计法（Normalized Statistics,NS）.该方法需要首先分析DAA协议中的各种复杂运算,针对不同的运算选用当前性能较好的算法,然后统计各个算法中大整数单精度乘法、单精度加法、读内存、写内存等基本运算的数目,最后通过汇总并转换得出DAA协议中各实体以机器周期为单位的性能负荷分布和总性能负荷.比较分析表明,该方法不仅能相对准确、精细、有效地定量计算出DAA协议中各实体的性能负荷和总的性能负荷,而且测出的性能负荷具有平台无关性.最后为了说明该方法的有效性,将NS方法应用于有关可信计算匿名证明的一个典型方案的性能负荷估算.

In order to improve wind tunnel test conditions and increase test efficiency,the research on the 1.2m trans-supersonic wind tunnel operation monitoring is carried out.The wireless sensor network(WSN) is introduced to construct the multi-sensor distributed wind tunnel monitoring platform in accordance with the complex operation environment and decentralized monitoring objects.The WSN data fusion based on back propagation (BP) algorithm neural network and multi-sensor data fusion based on D-S evidence theory are adopted in the wind tunnel monitoring application.This work has brought satisfactory effects in practice.

Generalization performance of support vector machines (SVM) with Gaussian kernel is influenced by its model parameters, both the error penalty parameter and the Gaussian kernel parameter. After researching the characteristics and properties of the parameter simultaneous variation of support vector machines with Gaussian kernel by the parameter analysis table, a new area distribution model is proposed, which consists of optimal straight line, reference point of area boundary, optimal area, transition area, underfitting area, and overfitting area. In order to improve classification performance of support vector machines, a genetic algorithm based on change area search is proposed. Comparison experiments show that the test accuracy of the genetic algorithm based on change area search is better than that of the two-linear search method.

Rough set theory is an effective mathematical tool for dealing with inconsistencies in information systems. Dominance based rough set (DBRS) is an extension to the original rough set, in which the equivalence relation is replaced by a dominance relation. However, in some condition, the lower approximation of DBRS can be emptied by only one "malicious" object. The variable consistency dominance based rough set (VC-DBRS) is proposed to avoid situations like this. The core conception of VC-DBRS is introducing parameters to control the consistency of objects including in lower approximations. There are several kinds of consistency measures have been proposed, but it is difficult to compute them by manual, especially for a large data set. It is necessary to find out an approach to calculate these measures automatically. This paper proposes a new algorithm based on dominance matrices to calculate the rough membership for VC-DBRS, and then presents how to use this measure to get the lower and upper approximation.

In this paper, we propose a new type of adaptive weight based on the definiteness of rough clusters and a hybrid threshold by combining the difference and distance threshold. And then, we refine the algorithm for assigning objects based on the hybrid thresholds to ensure that the outliers in inline positions and rectangle positions to be represented reasonably. At last, some experiments are provided to compare this refined RCM with the original RCM.

The goal of this paper is to investigate the cloud computing in smaert grid.First of all,we introduced the basicconcept of cloud computing and analyzed its major characteristics.Secondly,we combined cloud computing with smart grid and presented the view of smart cloud.Subsequently,we looked deeper into the three key issues,including construction of smart cloud,resource management and security policy.Finally,we pointed out future directions of research and concluded the paper.

The concept of green computing has begun to spread in the past few years, gaining increasing popularity. Besides the widespread sensitivity to ecological issues, such interest also stems from economic needs, since both energy costs and electrical requirements of IT industry around the world show a continuously growing trend. In this paper, we first discuss the connotation of green computing and sketch our view on the next generation of IT systems for green computing. Subsequently, we identify key issues relevant to green computing and evaluate different approaches to these problems. Finally, we point out future directions of research and conclude the paper.

In sensor networks, the selection of the transmission power level influences the network topology, power consumption and packet reception rate. In this paper, we analyze the effects of power control in the scenario of body sensor networks. A joint power control strategy considering both the MAC layer and the routing layer is also proposed. Simulation results show it achieves very good performance on both reliability and power efficiency.

Support vector machines (SVM) are a popular classification technique in diverse application areas. In this paper, the asymptotic behaviors of SVM are fused with clonal selection algorithm in artificial immune system; the search of the algorithm is directed to the straight line of optimal generalization error in the superparameter space. By the generating feature antibodies operation, the design of the antibody coding and the design of the affinity function, a new clonal selection algorithm based on feature antibodies, termed CSAFA, is proposed to construct CSAFA-SVM hybrid system so as to simultaneously optimize the feature subset and the parameters for SVM. The experimental results indicate that the proposed algorithm has higher classification accuracy rate, smaller feature subset and better performance compared with the existing clonal selection algorithm and other classification methods. Copyright © 2011 Binary Information Press.

Since its emergence, smart grid has been given increasingly widespread attentions. Basically, smart grid combines a various modern technologies like network communication, information processing and distributed control to provide a more secure, reliable and intelligent grid, thus meeting the requirements of future social and economic development. As a new paradigm in the power grid, smart grid undoubtedly represents the mainstream trend of future electric industry. As a result, it also brings some new technical challenges to researchers and engineering practioners. To support researchers and engineering practioners constructing a modern and intelligent grid, research in the field of smart grid has proliferated. In this paper, we look deeper into some key issues of smart grid, such as distributed cooperation and control, data and application integration, and knowledge-based comprehensive decision. Still, we give some solutions to resolve these challenges. In addition, we also introduce the concept of smart grid and define its key characteristics. Finally, we outline future directions of research and conclude the paper.