Catálogo de publicaciones - libros

The protein structure alignment problem addresses the question of measuring the degree of similarity, in three–dimensional structure, of two proteins. The representation of each protein using a simple contact map allows the correspondence graph for the protein pair to be generated and the maximum clique within this graph provides a measure of the structural similarity between the two proteins. This study uses a recently developed maximum clique algorithm, Phased Local Search (PLS), to locate the maximum cliques within correspondence graphs.

An important skill considered part of intelligent behavior is Abstract Thinking and Decision Making which includes thinking about a problem and reaching a decision after reasoning about its different aspects. This aspect of intelligent behavior has not recieved much attention and most of the cognitive architectures present today either focus more on the perceptual-motoric aspects of human brain or delve into the psycological, common behavioral and common sense issues. Here we present a cognitive architecture which addresses the issue of Abstract Thinking and Decision Making by using a novel representation for knowledge in the memory of an agent. The memory of an agent consists of four components, Concept Net, Working Memory, Perceptions and Possessions. Concept Net is a multi-layered net representing various concepts and their relationships with one another. We contend that this way of knowledge representation supports the process of decision making.

This paper presents a genetic algorithm (GA) based segmentation technique that can separate two touching objects intended for an automatic recognition of plastic bottles moving on a conveyor belt. The proposed method is based on the possibility to separate the two objects by means of a straight line, whose position is determined by a GA. Extensive testing shows that the proposed method is fast and yields high success rate of correct segmentation with only a limited number of both chromosomes and iterations.

Even prior to content, the genre of a web document leads to a first coarse binary classification of the recall space in relevant and non-relevant documents. Thinking of a genre search engine, massive data will be available via explicit or implicit user feedback. These data can be used to improve and to customize the underlying classifiers. A taxonomy of user behaviors is applied to model different scenarios of information gain. Elements of such a learning interface, as for example the implications of the and the , are discussed.

This paper presents a system which acquires noun concepts (labels for images) based on infant vocabulary acquisition models. In order to improve its performance, we introduced ”label acquisition rules” into this system.

Factor analysis is a statistical technique for reducing the number of factors responsible for a matrix of correlations to a smaller number of factors that may reflect underlying variables. Earlier experiments with constraint satisfaction problems (CSPs) using factor analysis suggested that there are only a few distinct principles of heuristic action. Here, this work is extended to the analysis of branching rules for SAT problems using the Davis-Putnam algorithm. These experiments show that just as with CSPs, there seem to be two basic actions that distinguish heuristics, characterised as building up of contention and propagation of effects to the uninstantiated portion of the problem.

One of the key applications of microarray studies is to select and classify gene expression profiles of cancer and normal subjects. In this study, two hybrid approaches–genetic algorithm with decision tree (GADT) and genetic algorithm with neural network (GANN)–are utilized to select optimal gene sets which contribute to the highest classification accuracy. Two benchmark microarray datasets were tested, and the most significant disease related genes have been identified. Furthermore, the selected gene sets achieved comparably high sample classification accuracy (96.79% and 94.92% in colon cancer dataset, 98.67% and 98.05% in leukemia dataset) compared with those obtained by mRMR algorithm. The study results indicate that these two hybrid methods are able to select disease related genes and improve classification accuracy.

A maximum likelihood blind source separation algorithm is developed. The temporal dependencies are explained by assuming that each source is an autoregressive (AR) process and the distribution of the associated i.i.d. innovations process is described by generalized Gaussian distribution (GGD), which can fit a broader range of statistical distributions by varying the value of the steepness parameter . Unlike most maximum likelihood methods the proposed algorithm takes into account both spatial and temporal information. Optimization is performed using the Expectation-Maximization method, and the source model is learned alone with the demixing parameters.

There has been much progresses recently about the identification of diagnostic proteomic signatures for different human cancers using surface-enhanced laser desorption ionization time-of-flight (SELDI-TOF) mass spectrometry. To identify proteomic patterns in serum to discriminate cancer patients from normal individuals, many classification methods have been experimented, often with successful results. Most of these earlier studies, however, are based on the direct application of original mass spectra, together with dimension reduction methods like PCA or feature selection methods like T-tests. Because only the peaks of MS data correspond to potential biomarkers, it is important to study classification methods using the detected peaks. This paper investigates ovarian cancer identification from the detected MS peaks by applying Bagging Support Vector Machine as a special strategy of bootstrap aggregating (Bagging). In bagging SVM, each individual SVM is trained independently, using randomly chosen training samples via a bootstrap technique. The trained individual SVMs are aggregated to make a collective decision in an appropriate way, for example, the majority voting. Bagged SVM demonstrated a 94% accuracy with 95% sensitivity and 92% specificity respectively by using the detected peaks. The efficiency can be further improved by applying PCA to reduce the dimension.

In this paper, we propose a novel framework to deal with data imbalance in class association rule mining. In each class association rule, the right-hand is a target class while the left-hand may contain one or more attributes. This framework is focused on the multiple imbalanced attributes on the left-hand. In the proposed framework, the rules with and without imbalanced attributes are processed in parallel. The rules without imbalanced attributes are mined through standard algorithm while the rules with imbalanced attributes are mined based on new defined measurements. Through simple transformation, these measurements can be in a uniform space so that only a few parameters need to be specified by user. In the case study, the proposed algorithm is applied into social security field. Although some attributes are severely imbalanced, the rules with minority of the imbalanced attributes have been mined efficiently.