Catálogo de publicaciones - libros

Composed of many classes or modules, big software can be represented with network model. By extracting the topology of UML metamodel from the UML metamodel specification, the scale-free, small-world networks properties are revealed. Based on this observation, we come up with our algorithms that can classify all classes in UML metamodel into three kinds: core, general and leaf. Our algorithm can categorize all classes into several subgroups by three factors, i.e., degree, betweenness and weak link. It is illustrated through case study that this algorithm is effective at mining community structure in large software systems.

Network intrusion detection is an important technique in computer security. However, the performance of existing intrusion detection systems (IDSs) is unsatisfactory since new attacks are constantly developed and the speed of network traffic volumes increases fast. To improve the performance of IDSs both in accuracy and speed, this paper proposes a novel adaptive intrusion detection method based on principal component analysis (PCA) and support vector machines (SVMs). By making use of PCA, the dimension of network data patterns is reduced significantly. The multi-class SVMs are employed to construct classification models based on training data processed by PCA. Due to the generalization ability of SVMs, the proposed method has good classification performance without tedious parameter tuning. Dimension reduction using PCA may improve accuracy further. The method is also superior to SVMs without PCA in fast training and detection speed. Experimental results on KDD-Cup99 intrusion detection data illustrate the effectiveness of the proposed method.

The infrastructure of grid information service constituted with highly distributed information providers and aggregate directory is brought forward on the basis of the characteristic of grid information resources in this paper. (LDAP), one of the base protocols, is also analyzed in this paper. It is put forward that LDAP is a distributed database. The dynamic updating and replication of LDAP directory tree happens frequently. To solve the problem, it has been proposed that the strategy of and can boost the efficiency of grid information service system. Moreover, we use file-parted replication approach to divide the LDAP database file into several blocks that are replicated parallel between LDAP sever points then. In such a way, the system efficiency of parallel processing can be boosted by margin. In addition, based on the idea forenamed, we put forward the technique infrastructure and , both of which are proven to be effective in improving the system efficiency.

The shape characteristic of daily power consumption of consumers can be applied to guide their power consumption behaviors and improve load structures of power system. It is also the basis to obtain the shape characteristic of daily power consumption of a trade and conduct researches in the state estimate of distribution networks etc. Traditional analytical approaches are limited to qualitative analysis with a small coverage only. We propose a model which can perform in-depth analysis of customer power consumption behaviors by data mining through similar sequence analysis to overcome the drawbacks of traditional approaches. The model uses real-time sampling of the energy data of consumers to form the shape characteristic curves. The application and testing of the model under an instance is analyzed in this paper.

Combining the advantages of the network and virtual SAN technology, the paper focuses on the characteristics of spatial data storage, by proposing a virtual SAN-based architecture of grid GIS data storage. Meanwhile, its corresponding experimental system has been designed to verify this proposal. In order for the storing facilities, FC and iSCSI data based double-thoroughfare have been designed in the proposed system to fulfill the efficient storage and retrieval of huge amount of spatial data. This solution possesses certain practical and applicable values from realizing and serving a technological foundation for exposable access and open interoperability of spatial data.

Desulfurization of molten iron is one of the stages of steel production process. A back-propagation (BP) artificial neural network (ANN) model is developed to predict the operation parameters for desulfurization process in this paper. The primary objective of the BP neural network predictor model is to assign the operation parameters on the basis of intelligent algorithm instead of the experience of operators. This paper presents a mathematical model and development methodology for predicting the three main operation parameters and optimizing the consumption of desulfurizer. Furthermore, a software package is developed based on this BP ANN predictor model. Finally, the feasibility of using neural networks to model the complex relationship between the parameters is been investigated.

This paper proposes flexible report architecture based on association rules data mining. A three-layer architecture is proposed namely, origin-data layer, data-processing layer, and format layer. These three layers are linked by a data variant tree in a power information management system. Users can modify report format as well as data whenever needed. In the origin-data layer data warehouse is used to provide data from multiple databases. In the data-processing layer, on-line analytical processing (OLAP) and association rules are used to enhance the template-making for reports. A smart solution to the problem of fixed report templates is provided and information in a power information management system can be shared. In some sense it can be an all-purpose tool to generate reports with great flexibility.

Privacy preserving data mining is to discover accurate patterns without precise access to the original data. In this paper, we combine the two strategies of data transform and data hiding to propose a new randomization method, Randomized Response with Partial Hiding (RRPH), for distorting the original data. Then, an effective naive Bayes classifier is presented to predict the class labels for unknown samples according to the distorted data by RRPH. Shown in the analytical and experimental results, our method can obtain significant improvements in terms of privacy, accuracy, and applicability.

Frequent itemset mining is a common task in data mining from which association rules are derived. As the frequent itemsets can be considered as a kind of summary of the original databases, recently the inverse frequent set mining problem has received more attention because of its potential threat to the privacy of the original dataset. Since this inverse problem has been proven to be NP-complete, people ask “ [1]. This paper describes our effort towards finding a feasible solution to address this problem.

Classification is one of the core topics in data mining technologies. This paper studies the geometry of classifying convex clusters based on support functionals in the dual spaces. For the convex clusters that are to be classified, a combination of linear discriminant functions could solve the problem. The geometrical depiction of linear discriminant functions and supporting hyperplanes for the convex clusters help to characterize the relations of the convex clusters, and the distances to the convex clusters and complement of convex clusters calibrate the measures between the support functionals and convex clusters. Examples are given.