Catálogo de publicaciones - libros

Along with the increasing amounts of XML data available, the data mining community has been motivated to discover the useful information from the collections of XML documents. One of the most popular approaches to find the information is to extract frequent subtrees from a set of XML trees. In this paper, we propose a novel algorithm, EXiT-B, for efficiently extracting maximal frequent subtrees from a set of XML documents. The main contribution of our algorithm is that there is no need to perform tree join operation during the phase of generating maximal frequent subtrees. Thus, the task of finding maximal frequent subtrees can be significantly simplified comparing to the previous approaches.

The application of Ontologies for the definition and interoperability of complementary taxonomies has been well-recognised within the Modelling & Simulation (M&S) community. Our research pertaining to the specification of (SE) representational semantics has proposed the use of an (), which is modeled using W3C’s (OWL). The vocabulary specified in Onto is based the SEDRIS Data Representation Model (DRM), which is a technological framework for SE data interchange and interoperability.

In this paper, we present STOWL – that automates the transformation of a SEDRIS based SE to a Web-Ontology based representation scheme in the OWL language. The target representation scheme, which shall be based on Onto, is in actuality an instantiation of the SE data representation terminology as specified by . Such a transformation has many perceived advantages: It enhances SE interoperability by utilizing a Web-Ontology based approach for the specification of SE representation data, is consistent with existing industry based SE representation standards, namely SEDRIS, and that the representation scheme facilitates ontological reasoning over SE objects; a facility that is not directly supported by the SEDRIS DRM.

In this article, we investigate the problem of checking consistency in a hybrid formalism, which combines two essential formalisms in qualitative spatial reasoning: topological formalism and cardinal direction formalism. Instead of using conventional composition tables, we investigate the interactions between topological and cardinal directional relations with the aid of rules that are used efficiently in many research fields such as content-based image retrieval. These rules are shown to be sound, i.e. the deductions are logically correct. Based on these rules, an improved constraint propagation algorithm is introduced to enforce the path consistency.

Many applications using sensing data require the fast retrieval of aggregated information in sensor networks. In this paper, distributed spatial index structure in sensor networks for time-efficient aggregation query processing is proposed. The main idea is to logically organize sensors in underlying networks into distributed R-Tree structure, named Sensor Tree. Each node of the Sensor Tree has pre-aggregated results which are the collection of the values of aggregated result for sensing data of the same type of sensors within Minimum Bounding Rectangle (MBR). If a spatial region query is required, the processing of the query searches the location of the target sensor from the root of the Sensor Tree. And then it finally sends the values of pre-aggregated result of that sensor. By the proposed Sensor Tree aggregation query processing on any region, response time and energy consumption can be reduced since it avoids flooding query to the leaf sensor or non-relevant sensors in the sensor networks.

In the task of classification, most learning methods are suitable only for certain data types. For the hybrid dataset consists of nominal and numeric attributes, to apply the learning algorithms, some attributes must be transformed into the appropriate types. This procedure could damage the nature of dataset. We propose a model tree approach to integrate several characteristically different learning methods to solve the classification problem. We employ the decision tree as the classification framework and incorporate support vector machines into the tree construction process. This design removes the discretization procedure usually necessary for tree construction while decision tree induction itself can deal with nominal attributes which may not be handled well by e.g., SVM methods. Experiments show that our purposed method has better performance than that of other competing learning methods.

For statistical modelling of multivariate binary data, such as text documents, datum instances are typically represented as vectors over a global vocabulary of attributes. Apart from the issue of high dimensionality, this also faces us with the problem of uneven importance of various attribute presences/absences. This problem has been largely overlooked in the literature, however it may create difficulties in obtaining reliable estimates of unsupervised probabilistic representation models. In turn, the problem of automated feature selection and feature weighting in the context of unsupervised learning is challenging, because there is no known target to guide the search. In this paper we propose and study a relatively simple cluster-based generative model for multivariate binary data, equipped with automated feature weighting capability. Empirical results on both synthetic and real data sets are given and discussed.

Knowledge reduction is one of the basic contents in rough set theory and one of the most problem in knowledge acquisition. The main objective of this paper is to introduce a new concept of knowledge reduction based on partition. It is referred to as partition reduction. The partition reduction is to unify the definitions of classical knowledge reductions. Classical knowledge reductions such as absolute attribute reduction, relative reduction, distribution reduction, assignment reduction and maximum distribution reduction are special cases of partition reduction. We can establish new types of knowledge reduction to meet our requirements based on partition reduction.

Multiresolution histograms have been used for indexing and retrieval of images. Multiresolution histograms used traditionally are 2d-histograms which encode pixel intensities. Earlier we proposed a method for decomposing images by . In this paper, we propose to encode centroidal distances of an image in multiresolution histograms; the image is decomposed a priori, by connectivity. Multiresolution histograms thus obtained are 3d-histograms which encode connectivity and centroidal distances. The statistical technique of Principal Component Analysis is applied to multiresolution 3d-histograms and the resulting data is used to index images. Distance between two images is computed as the -difference of their principal components. Experiments are performed on Item S8 within the MPEG-7 image dataset. We also analyse the effect of pixel intensity thresholding on multiresolution images.

It is known that no single descriptor is powerful enough to encompass all aspects of image content, i.e. each feature extraction method has its own view of the image content. A possible approach to cope with that fact is to get a whole view of the image(object). Then using machine learning approach from user’s Relevance feedback to obtain a reduced feature. In this paper, we concentrate on some points about Biased Discriminant Analysis / Kernel Biased Discriminant Analysis (BDA/KBDA) based machine learning approach for CBIR. The contributions of this paper are: 1. using generalized singular value decomposition (GSVD) based approach solve the small sample size problem in BDA/KBDA and 2. using histogram intersection as a kernel for KBDA. Experiments show that this kind of kernel gets improvement compare to other common kernels.

Fuzzy c-means (FCM) clustering algorithm as an unsupervised fuzzy clustering technique has been widely used in image segmentation. However, the conventional FCM algorithm is very sensitive to noise for the reason of incorporating no information about spatial context while segmentation. To overcome this limitation of FCM algorithm, a novel penalized fuzzy c-means (PFCM) algorithm for image segmentation is presented in this paper. The algorithm is realized by incorporating the spatial neighborhood information into the original FCM algorithm by a penalty term, which is employed to take into account the spatial dependence of the objects. Experiments demonstrate the proposed algorithm is effective and more robust to noise and other artifacts than the standard FCM algorithm.