Catálogo de publicaciones - libros

Automobile panel surface is a kind of complicated 3-D free-from surface. According to the characteristic of the shape of the panel surface, a surface recognition scheme describing the surface in the primary inertia axes coordinate system is proposed. In the coordinate system, the surface equation will keep invariant when the surface is translated or rotated. After extracting the surface’s outline projection curve and section curves, the curve’s eigenvectors are calculated with an improved invariant moment method. At last, the similarity between two surfaces is figured out based on the similarity between the section curves extracted from the two surfaces separately.

Artificial neural network can be categorized according to the type of learning, that is, supervised learning versus unsupervised learning. Unsupervised learning can find the major features of the origin data without indication. Adaptive resonance theory can classify large various data into groups of patterns. Through analysing the limit of adaptive resonance theory, a dynamic clustering algorithm is provided. The algorithm not only can prevent from discarding irregular data or giving rise to dead neurons but also can cluster unlabelled data when the number of clustering is unknown. In the experiments, the same data are used to train the adaptive resonance theory network and the dynamic clustering algorithm network. The results prove that dynamic clustering algorithm can cluster unlabelled data correctly.

Current Semantic Web community has popularized ontology research. However, ontology building by hand has proven to be a very hard and error-prone task and become the bottleneck of ontology acquiring process. WordNet, an electronic lexical database, is considered to be the most important resource available to researchers in computational linguistics. The paper proposes an ontology learning approach, which uses WordNet lexicon resources to build a standard OWL ontology model. The approach will help the automation of ontology building and be very useful in ontology-based applications.

Phylogeny reconstruction is a difficult computational problem, because the number of possible solutions increases with the number of included taxa. For this reason, phylogenetic inference methods commonly use clustering algorithms or heuristic search strategies to minimize the amount of time spent evaluating non-optimal trees. Even heuristic searches can be painfully slow, especially when computationally intensive optimality criteria such as maximum likelihood are used. I describe here a genetic programming to heuristic searching that can tremendously reduce the time required for maximum-likelihood phylogenetic inference, especially for data sets involving large numbers of taxa, and we confirm its availability by experiments.

In general, there is a preference order on the domain of an attribute in preference-ordered data, but most data mining approaches ignore it. Such attribute is a criterion. In fact, such rules are useful for prediction: if the mathematics score of a student goes up then his/her physics score will go up. Such rules are called dominance association rules here. Based on criteria, a preference-ordered data table can be transformed into a tri-value data table, in which dominance association rules can be mined. Dominance association rules uncover the correlation between criteria and reflect when values of a set of criteria change, how values of another set of criteria change. One use of dominance association rules is to predict the unknown values of criteria in an object by comparing with other objects.

It is well recognized that sequential pattern mining plays an essential role in many scientific and business domains. In this paper, a new extension of sequential pattern, ordinal pattern, is proposed. An ordinal pattern is an ordinal sequence of attributes, whose values commonly occur in ascending order over data set. After each record in data set is transformed into an ordinal sequence of attributes according to their ordinal values, ordinal patterns can be mined by means of mining sequential patterns. One use of ordinal patterns is to identify possible error records in data cleaning, in which the values of attributes are inconsistent with the ordinal patterns which most of the data conform to. Experiments verify the high efficiency of the method presented.

The problem of mining user association from rating table plays an essential role in rule-based recommender system. Using the closure of the Galois connection, we define two user association bases: the exact base (i.e., for all rules with a 100% confidence) and the approximate base (i.e., with confidence <100%) from which all valid user association rules with support and confidence can deduced. These user association bases are characterized using frequent closed itemsets and their reductions within concept lattice. Algorithm for extracting these two bases is presented and experimental evaluated on real-life databases. The results show that the proposed user association bases can considerably reduce number of rules in user association and do not loss any information.

It is a significant thing that generating specified texture picture in specified position. This paper proposed a novel texture synthesis method based upon multi-sample of texture. particle swarm optimization was applied to carry out single-sample texture synthesis, and then it was applied to constrained multi-sample texture synthesis in extension. The users can generate new pictures according to their own demands. Experiment shows that the algorithm is satisfied in the aspect of speed and quality.

Data mining, i.e., clustering analysis, is a challenging task due to the huge amounts of data. In this paper, we propose a general incremental hierarchical clustering method dealing with incremental data sets in data warehouse environment for data mining to reduce the cost further. As an example, we put forward ICHAMELEON, the improvement of CHAMELEON, which is a hierarchical clustering method, and demonstrate that ICHAMELEON is highly efficient in terms of time complexity. Experimental results on very large data sets are presented which show the efficiency of ICHAMELEON compared with CHAMELEON.

OLAP (on-line analytical processing) systems support decision-making process by providing dynamic analytical operations on high volumes of data. Usually, the operations require dimensions to be regular, however, in real-world applications, many complex dimensions fail to meet the requirement. In this paper, we first propose a new conceptual model, compared with traditional multidimensional models, this model extends the surjection between the domains of two levels to the partial mapping. Afterwards, we present a transforming algorithm to prove that the model offers a practical way for handling irregular dimensions in OLAP systems.