Search

Application of kernel methods to link analysis is presented. Novel kernels based on directed graph Laplacians are proposed and their application as measures of relatedness between nodes in a directed graph is presented. The kernels express relatedness and take into account the global importance of the nodes in a citation graph. Limitations of existing kernels are given with a discussion how they are addressed by directed Laplacian kernels. Links between the kernels and PageRank ranking algorithm are also presented. The proposed kernels are evaluated on a dataset of scientific bibliographic citations. [ABSTRACT FROM AUTHOR]

Many applications require ordering of instances represented by high dimensional vectors. Despite the reasonable quantity of papers on classification and clustering, papers on multidimensional ranking are rare. This paper expands a generic ranking procedure based on one-dimensional self-organizing maps (SOMs). The typical similarity metric is modified to a weighted Euclidean metric and automatically adjusted by a genetic search. The search goal is the best ranking that matches the desired probability distribution (provided by experts) leading to a context-sensitive metric. To ease expert agreement the technique relies on consensus about the best and worst instances. Besides the ranking task, the derived metric is also useful on reducing the number of dimensions (questionnaire items in some situations) and on modeling the data source. Promising results were achieved on the ranking of data from blood bank inspections and client segmentation in agribusiness. [ABSTRACT FROM AUTHOR]

This paper proposes markovian models in portfolio theory and risk management. In a first analysis, we describe discrete time optimal allocation models. Then, we examine the investor's optimal choices either when returns are uniquely determined by their mean and variance or when they are modeled by a Markov chain. Moreover we propose different models to compute VaR and CVaR when returns are modeled by a Markov chain. [ABSTRACT FROM AUTHOR]

This paper proposes a Flexible Neural Tree (FNT) model for informative gene selection and gene expression profiles classification. Based on the pre-defined instruction/operator sets, a flexible neural tree model can be created and evolved. This framework allows input variables selection, over-layer connections and different activation functions for the various nodes involved. The FNT structure is developed using the Extended Compact Genetic Programming and the free parameters embedded in the neural tree are optimized by particle swarm optimization algorithm. Empirical results on two well-known cancer datasets shows competitive results with existing methods. [ABSTRACT FROM AUTHOR]

This paper proposes an efficient method for indexing and mining graph database. Most existing approaches are based on frequent sub-structures such as edges, paths, or subgraphs. However, as the size of graphs increases, such index structure grows drastically in size for avoiding performance degradation. This yields a requirement for constructing a more compact index structure and introducing more informative indexing items into this index to increase its pruning power. In this paper, we demonstrate that degree information can help solve this problem. Based on this idea, we propose a new index structure (D-index) which uses the subgraph and its degree information as the indexing item. Our empirical study shows that D-index achieves remarkable improvement in performance over the state-of-the-art approach. [ABSTRACT FROM AUTHOR]

This paper tries to propose the worm virus detection system that focuses on many connection attempts, more frequently occurring in the process of scanning than their common transmission processes. And this paper tries to determine the critical value of connection attempt by using the ordinary time network traffic learning technique which applies the genetic algorithm in order to ensure accurate detection of virus, depending on the status of network. This system can reduce the damage from worm virus more quickly than the pattern-founded worm virus detection system because it applies the common characteristics of worm viruses to detect them, and the criteria for judgment can be altered in its application though the network may change. [ABSTRACT FROM AUTHOR]

Motion segmentation needs to estimate the parameters of motion and its supporting region. The usual problem in determining the supporting region is how to obtain a complete spatial consistence. On the basis of maximum posterior marginal probability (MPM-MAP) algorithm this paper presents a new algorithm based on region shrinking to locate the supporting area. First the motion parameters are estimated by MPM-MAP algorithm. In this algorithm pixels of maximum probabilities belonging to a motion are considered to be preselected pixels for supporting area. Then the region shrinking algorithm is used to determine the region of maximum density of the preselected pixels to be the range of supporting area. Finally the active contour based on gradient vector flow (GVF) is adopted to obtain the accurate shape of supporting region. This method obtains a solid region to be supporting area of a motion and extracts the accurate shape of moving objects, so it offers a better way in motion segmentation to solve the problem of spatial continuity. Keywords: region shrinking; motion segmentation; MPM-MAP; active contour; gradient vector flow. [ABSTRACT FROM AUTHOR]

Statistical re-sampling techniques have been used extensively and successfully in the machine learning approaches for generations of classifier and predictor ensembles. It has been frequently shown that combining so called unstable predictors has a stabilizing effect on and improves the performance of the prediction system generated in this way. In this paper we use the re-sampling techniques in the context of a topology preserving map which can be used for scale invariant classification, taking into account the fact that it models the residual after feedback with a family of distributions and finds filters which make the residuals most likely under this model. This model is applied to artificial data sets and compared with a similar version based on the Self Organising Map (SOM). [ABSTRACT FROM AUTHOR]

Evolutionary algorithms has been recently used for prototype selection showing good results. An important problem in prototype selection consist in increasing the size of data sets. This problem can be harmful in evolutionary algorithms by deteriorating the convergence and increasing the time complexity. In this paper, we offer a preliminary proposal to solve these drawbacks. We propose an evolutionary algorithm that incorporates knowledge about the prototype selection problem. This study includes a comparison between our proposal and other evolutionary and non-evolutionary prototype selection algorithms. The results show that incorporating knowledge improves the performance of evolutionary algorithms and considerably reduces time execution. [ABSTRACT FROM AUTHOR]

D-S evidence theory has been studied and used for information fusion for a while. Though D-S evidence theory can deal with uncertainty reasoning from imprecise and uncertain information by combining cumulative evidences for changing prior opinions using new evidences. False evidence generated by any fault sensor will result in evidence conflict increasing and inaccurate fused results. Evidence relationship matrix proposed in this paper depicts the relationship among evidences. False evidences can be identified through the analysis of relationships among evidences. Basic probability assignments related to the false evidences may be decreased accordingly. The accuracy of information fusion may be improved. Case studies show the effectiveness of the proposed method. Keywords: evidence relationship matrix, D-S evidence theory, and information fusion. [ABSTRACT FROM AUTHOR]

This paper presents new application of Particle Swarm Optimization (PSO) algorithm for training Hidden Markov Models (HMMs). The problem of finding an optimal set of model parameters is numerical optimization problem constrained by stochastic character of HMM parameters. Constraint handling is carried out using three different ways and the results are compared to Baum-Welch algorithm (BW), commonly used for HMM training. The global searching PSO method is much less sensitive to local extremes and finds better solutions than the local BW algorithm, which often converges to local optima. The advantage of PSO approach was markedly evident, when longer training sequence was used. [ABSTRACT FROM AUTHOR]

This paper analysis dynamics of JPY/USD and USD/EUR currency option RND (Risk-Neutral Density) function and moments during CNY realignment. Empirical study shows CNY realignment only has limited impact on USD expectation. The influence of CNY on FX market should not be exaggerated. Keywords: CNY realignment, USD expectation, Currency option, RND function. [ABSTRACT FROM AUTHOR]

In this paper, first, our J-IFC-BRIDGE product model, which is a combination of YLPC-BRIDGE and YLSG-BRIDGE, and French IFC-BRIDGE product models were described. Then, the process of merging these models into the New IFC-BRIDGE was presented, and New IFC-BRIDGE was introduced. Further, the concrete bridge design system using multi-agents that had been developed before was modified to follow the changes of the product model, and the new system and the product model were verified by applying the design scenario of a prestressed concrete girder. Both the New IFC-BRIDGE and the concrete bridge design system showed the feasibility and practicality. Keywords: product data model, IFC, bridge, design checking, multi-agents. [ABSTRACT FROM AUTHOR]

This paper presents a multiagent model that facilitates aspects of shopping mall management, as well as increasing the quality of leisure facilities and shopping on offer. The work presented focuses on the use of a multi agent architecture, based on the use of deliberative agents that incorporates case-based planning. The architecture considers a dynamic framework, and the need to use autonomous models that are able to evolve over time. The architecture incorporates agents whose aim is to acquire knowledge and adapt themselves to the environmental changes. The system has been tested successfully, and the results obtained are presented in this paper. Keywords: CBR, CBP-BDI, wireless multiagent system, shopping mall, SMAS. [ABSTRACT FROM AUTHOR]

Aimed at multimodal optimization, the paper proposes an improved discrete immune network model (IDIN). In order to mention dynamic population and appropriate diversity, we introduce a bi-shape-space coding mechanism into a discrete immune network. To evaluate the efficiency of IDIN, we implemented three benchmark functions with various dimensions, and compared our results with other algorithms based on immune algorithm or the developed version of Genetic Algorithms. We have obtained similar or better results than those of others. Keywords: Bi-shape-space coding; Discrete immune network; Benchmark functions. [ABSTRACT FROM AUTHOR]

In this paper we explore the effectiveness of combining diverse machine learning based methods to categorize patent applications. Classifiers are constructed from each categorization method in the combination, based on the document representations where the best performance was obtained. Therefore, the ensemble of methods makes categorization predictions with knowledge observed from different perspectives. In addition, we explore the application of a variety of combination techniques to improve the overall performance of the ensemble of classifiers. In our experiments a refined version of the WIPO-alphaWIPO-alpha document collection available at http://www.wipo.int/ibis/datasets/index.html document collection was used to train and evaluate the classifiers. The combination ensemble that achieved the best performance obtained an improvement of 6.51% compared to the best performing classifier participating in the combination. Keywords: Categorization, Machine Learning, Knowledge Management. [ABSTRACT FROM AUTHOR]

Globalization and deregulation are modifying the competitive framework in the majority of economic sectors and, as a result, many companies are changing their commercial model to focus on the preservation of existing customers. Understanding customer loyalty therefore represents an element of competitive advantage. In this brief paper, we investigate loyalty in the Spanish petrol station market, according to the customer satisfaction and switching barriers constructs. Satisfaction and behavioural intentions are analysed within a classification framework using Bayesian neural networks. The necessary interpretability and actionability of the results is achieved through the use of a feature selection process embedded in the network training and a novel rule extraction method. [ABSTRACT FROM AUTHOR]

In this paper, we focus on the problem of feature selection with confidence machines (CM). CM allows us to make predictions within predefined confidence levels, thus providing a controlled and calibrated classification environment. We present a new feature selection method, namely Strangeness Minimisation Feature Selection, designed for CM. We apply this feature selection method to the problem of microarray classification and demonstrate its effectiveness. [ABSTRACT FROM AUTHOR]

The mass function of evidential theory provides a means of representing ignorance in lack of information. In this paper we propose mass function models of aggregate views held as summary tables in a distributed database. This model particularly suits statistical databases in which the data usually presents imprecision, including missing values and overlapped categories of aggregate classification. A new aggregation combination operator is developed to accomplish the integration of semantically heterogeneous aggregate views in such distributed databases. [ABSTRACT FROM AUTHOR]

In this paper we apply Biased Minimax Probability Machine (BMPM) to address the problem of relevance feedback in Content-based Image Retrieval (CBIR). In our proposed methodology we treat relevance feedback task in CBIR as an imbalanced learning task which is more reasonable than traditional methods since the negative instances largely outnumber the positive instances. Furthermore we incorporate active learning in order to improve the framework performance, i.e., try to reduce the number of iterations used to achieve the optimal boundary between relevant and irrelevant images. Different from previous works, this model builds up a biased classifier and achieves the optimal boundary using fewer iterations. Experiments are performed to evaluate the efficiency of our method, and promising experimental results are obtained. [ABSTRACT FROM AUTHOR]

This paper describes a knowledge-based approach for summarizing and presenting the behavior of hydrologic networks. This approach has been designed for visualizing data from sensors and simulations in the context of emergencies caused by floods. It follows a solution for event summarization that exploits physical properties of the dynamic system to automatically generate summaries of relevant data. The summarized information is presented using different modes such as text, 2D graphics and 3D animations on virtual terrains. The presentation is automatically generated using a hierarchical planner with abstract presentation fragments corresponding to discourse patterns, taking into account the characteristics of the user who receives the information and constraints imposed by the communication devices (mobile phone, computer, fax, etc.). An application following this approach has been developed for a national hydrologic information infrastructure of Spain. [ABSTRACT FROM AUTHOR]

With the increasing expansion of network scale, users consume much cost for network that is composed of various network environments. In this paper, we propose a dynamic network management system employing the management program actively generating facilities in order to reduce the cost and time of system development. The proposed system consists of basic components for real time network management and Automatic Generation Technique of Network Components (AG-NC) in order to actively support information generation for network management objects through these basic components. The basic components includes configuration management, performance management, and fault management. AG-NC includes NE Basic Info Handler, MIB Handler, Template Handler, and Operation Handler. This system is able to generate a network management program with SNMP manager using the information of network objects. Our system actively generates network objects in an automatic manner instead of the manual manner of the existing systems. Finally, we can make the network structure expansion because the development time and cost of the network management program can be reduced dramatically through our system. [ABSTRACT FROM AUTHOR]

This paper presents some of our first results in the development of an automatic sound classification algorithm for hearing aids. The goal is to classify the input audio signal into four different categories: speech in quiet, speech in noise, stationary noise and non-stationary noise. In order to make the system more robust, a divide and conquer strategy is proposed, resulting thus in a layered structure. The considered classification algorithms will be based on the Fisher linear discriminant and neural networks. Some results will be given demonstrating the good behavior of the system compared with a classical approach with a four-classes classifier based on neural networks. [ABSTRACT FROM AUTHOR]

The problem of similarity search in time series database has attracted a lot of interest in the data mining field. DTW(Dynamic Time Warping) is a robust distance measure function for time series, which can handle time shifting and scaling. The main defect of DTW lies in its relatively high computational complexity of similarity search. In this paper, we develop a simple but efficient approximation technique for DTW to speed up the search process. Our method is based on a variation of the traditional histograms of the time series. This method can work with a time linear with the size of the database. In our experiment, we proved that the proposed technique is efficient and produces few false dismissals in most applications. [ABSTRACT FROM AUTHOR]

Attribute reduction is an important issue of data mining. It is generally regarded as a preprocessing phase that alleviates the curse of dimensionality, though it also leads to classificatory analysis of decision tables. In this paper, we propose an efficient algorithm TWI-SQUEEZE that can find a minimal (or irreducible) attribute subset, which preserves classificatory consistency after two scans of a decision table. Its worst-case computational complexity is analyzed. The outputs of the algorithm are two different kinds of classifiers. One is an IF-THEN rule system. The other is a decision tree. [ABSTRACT FROM AUTHOR]

Parametric software effort estimation models rely on the availability of historical project databases from which estimation models are derived. In the case of large project databases with data coming from heterogeneous sources, a single mathematical model cannot properly capture the diverse nature of the projects under consideration. Clustering algorithms can be used to segment the project database, obtaining several segmented models. In this paper, a new tool is presented, Recursive Clustering Tool, which implements the EM algorithm to cluster the projects, and allows use different regression curves to fit the different segmented models. This different approaches will be compared to each other and with respect to the parametric model that is not segmented. The results allows conclude that depending on the arrangement and characteristics of the given clusters, one regression approach or another must be used, and in general, the segmented model improve the unsegmented one. Keywords: Software Engineering, Effort estimation, Segmented parametric model, Recursive Clustering Tool (RCT), Clustering, EM algorithm. [ABSTRACT FROM AUTHOR]

Since cluster analysis in data mining often deals with large-scale high-dimensional data with masking variables, it is important to remove non-contributing variables for accurate cluster recovery and also for proper interpretation of clustering results. Although the weights obtained by variable weighting methods can be used for the purpose of variable selection (or, elimination), they alone hardly provide a clear guide on selecting variables for subsequent analysis. In addition, variable selection and variable weighting are highly interrelated with the choice on the number of clusters. In this paper, we propose a non-parametric data clustering method, based on the W-k-means type clustering, for an automated and joint decision on selecting variables, determining variable weights, and deciding the number of clusters. Conclusions are drawn from computational experiments with random data and real-life data. [ABSTRACT FROM AUTHOR]

In the knowledge economy taxonomy generation, information retrieval and portals in intelligent enterprises need to be dynamically adaptive to changes in their enterprise content. To remain competitive and efficient, this has to be done without exclusively relying on knowledge workers to update taxonomies or manually label documents. This paper briefly reviews existing visualisation methods used in presenting search results retrieved from a web search engine. A method, termed topological tree, that could be use to automatically organise large sets of documents retrieved from any type of search, is presented. The retrieved results, organised using an online version of the topological tree method, are compared to the visual representation of a web search engine that uses a document clustering algorithm. A discussion is made on the criterions of representing hierarchical relationships, having visual scalability, presenting underlying topics extracted from the document set, and providing a clear view of the connections between topics. The topological tree has been found to be a superior representation in all cases and well suited for organising web content. Keywords: Information retrieval, document clustering, search engine, self organizing maps, topological tree, information access, faceted classification, guided navigation, taxonomy generation, neural networks, post retrieval clustering, taxonomy generation, enterprise portals, enterprise content management, enterprise search, information management. [ABSTRACT FROM AUTHOR]

Combining audio and image processing for understanding video content has several benefits when compared to using each modality on their own. For the task of context and activity recognition in video sequences, it is important to explore both data streams to gather relevant information. In this paper we describe a video context and activity recognition model. Our work extracts a range of audio and visual features, followed by feature reduction and information fusion. We show that combining audio with video based decision making improves the quality of context and activity recognition in videos by 4% over audio data and 18% over image data. [ABSTRACT FROM AUTHOR]

This paper describes a new method to automatically obtain a new thesaurus which exploits previously collected information. Our method relies on different resources, such as a text collection, a set of source thesauri and other linguistic resources. We have applied different techniques in the different phases of the process. By applying indexing techniques, the text collection provides the set of initial terms of interest for the new thesaurus. Then, these terms are searched in the source thesauri, providing the initial structure of the new thesaurus. Finally, the new thesaurus is enriched by searching for new relationships among its terms. These relationships are first detected using similarity measures and then are characterized with a type (equivalence, hierarchy or associativity) by using different linguistic resources. We have based the system evaluation on the results obtained with and without the thesaurus in an information retrieval task proposed by the Cross-Language Evaluation Forum (CLEF). The results of these experiments have revealed a clear improvement of the performance. [ABSTRACT FROM AUTHOR]

The vision system of human beings is very sensitive to lines. Sketches composed of line drawings provide a useful representation for image structures, which can be used for recognition and which is capable of serving as intermediate data structures of images for further analysis. However, traditional schemes tackling the line detection problem for sketch generation are stuck to one of the two problems: lack of global supervision for line detection over an image or confinement of hard global constraints on the configuration of detected lines.In this paper, we propose an active sketch model for finding primary structures in the form of line drawings, overcoming shortcomings of the traditional schemes by combining local line finding techniques with supervision of a global saliency measure while imposing no global constraint on the configuration of the line drawings.Test results show that our model is able to sketch out lines in an image representing the most salient image discontinuity without imposing any shape or topology constraint on the sketch. [ABSTRACT FROM AUTHOR]

Using background knowledge in clustering, called semi-clustering, is one of the actively researched areas in data mining. In this paper, we illustrate how to use background knowledge related to a domain more efficiently. For a given data, the number of classes is investigated by using the must-link constraints before clustering and these must-link data are assigned to the corresponding classes. When the clustering algorithm is applied, we make use of the cannot-link constraints for assignment. The proposed clustering approach improves the result of COP k-means by about 10%. [ABSTRACT FROM AUTHOR]

Recently, as the size of genetic knowledge grows faster, the automated analysis and systemization into high-throughput database has become a hot issue. In bioinformatics area, one of the essential tasks is to recognize and identify genomic entities and discover their relations from various sources. Generally, biological literatures containing ambiguous entities, are laid by decision boundaries. The purpose of this paper is to design and implement a classification system for improving performance in identifying entity problems. The system is based on reinforcement training and post-processing method and supplemented by data mining algorithms to enhance its performance. For experiments, we add some intentional noises to training data for testing the robustness and stability. The result shows significantly improved stability on training errors. [ABSTRACT FROM AUTHOR]

Dimension reduction techniques are important preprocessing algorithms for high dimensional applications that reduce the noise keeping the main structure of the dataset. They have been successfully applied to a large variety of problems and particularly in text mining applications. However, the algorithms proposed in the literature often suffer from a low discriminant power due to its unsupervised nature and to the ‘curse of dimensionality'. Fortunately several search engines such as Yahoo provide a manually created classification of a subset of documents that may be exploited to overcome this problem. In this paper we propose a semi-supervised version of a PCA like algorithm for textual data analysis. The new method reduces the term space dimensionality taking advantage of this document classification. The proposed algorithm has been evaluated using a text mining problem and it outperforms well known unsupervised techniques. [ABSTRACT FROM AUTHOR]

This paper deals with a learning algorithm which combines two well known methods to generate ensemble diversity - error negative correlation and resampling. In this algorithm, a set of learners iteratively and synchronously improve their state considering information about the performance of a fixed number of other learners in the ensemble, to generate a sort of local negative correlation. Resampling allows the base algorithm to control the impact of highly influential data points which in turns can improve its generalization error. The resulting algorithm can be viewed as a generalization of bagging, where each learner no longer is independent but can be locally coupled with other learners. We will demonstrate our technique on two real data sets using neural networks ensembles. [ABSTRACT FROM AUTHOR]

In this paper, a computer interface for handicapped people is proposed, where input signals are given by eye movement of the handicapped people. Eye movement is detected by neural network (NN)-based texture classifier, which enables our system to be not obliged to constrained environment. To be robust the natural motion of a user, we first detect a user's face using skin-color information, and then detect her or his eyes using neural network (NN)-based texture classifier. After detection of eye movements, the tracking is performed using mean-shift algorithms. We use this eye-tracking system as an interface to control the surrounding system such as audio, TV, light, phone, and so on. The experimental results verify the feasibility and validity of the proposed eye-tracking system to be applicable as an interface for the handicapped people. [ABSTRACT FROM AUTHOR]

Distributions of very high dimensional data are, in most cases, not even, but skewed. For this reason, there can be more effective dimensions than others in partitioning a high dimensional data set. Effective dimensions can be used to partition the data set in more balanced way so that data are located in more evenly distributed. In this paper, we present schemes to select dimensions by which high dimensional data sets are partitioned for efficient similarity joins. Especially, in order to efficiently reduce the number of partition dimensions, we propose a novel scheme using diagonal dimensions compared with perpendicular dimensions. The experimental results show that the proposed schemes substantially improve the performance of the partition-based similarity joins in high dimensional data spaces. [ABSTRACT FROM AUTHOR]

The global k-means with similarity functions algorithm is an algorithm that allows working with qualitative and quantitative features (mixed data), but it involves a heavy computational cost. Therefore, in this paper, an algorithm that accelerates the global k-means with similarity functions algorithm without significantly affecting the quality of the solution is proposed. Our algorithm called fast global k-means with similarity functions algorithm is tested and compared against the k-means with similarity functions algorithm and the global k-means with similarity functions algorithm. [ABSTRACT FROM AUTHOR]

Automatic microfossil detection system allows to extract the position of the microfossils in a concentrate of mineral grains, speeding up the time required to analyze each sample. In this paper we study the use of Multilayer Perceptrons and Radial Basis Function Networks applied to the automatic microfossil teeth detection problem, focusing on the dependence of the performance with the size of the network, and with the size of the training set. The data used in the experiments are three images of concentrates with micromammal teeth from Somosaguas paleontological site, in Madrid (Spain). The obtained results demonstrate RBFNs perform better than MLPs in most of the considered cases, detecting most of the microfossil teeth in the images. [ABSTRACT FROM AUTHOR]

Security has become a main concern in corporate networks. Security tests are essential to identify vulnerabilities, but experts must analyze very large data and complex information. Unsupervised learning can help by clustering groups of devices with similar vulnerabilities. However an index to evaluate every solution should be calculated to demonstrate results validity. Also the value of the number of clusters should be tuned for every data set in order to find the best solution. This paper introduces SOM as a clustering method to evaluate complex and uncertain knowledge in Consensus, a distributed security system for vulnerability testing; it proposes new metrics to evaluate the cohesion of every cluster, and also the cohesion between clusters; it applies unsupervised algorithms and validity metrics to a security data set; and it presents a method to obtain the best number of clusters regarding these new cohesion metrics: Intracohesion and Intercohesion factors. Keywords: AI applications, Unsupervised Learning, Self-Organization Map, K-means, Network security. [ABSTRACT FROM AUTHOR]

In this paper a method for learning and representing joint probabilistic distributions, using binary trees, is shown. This method could be used with the Bayesian Programming formalism, being a very useful tool when working with real world data. It has the advantage of learning unknown probabilistic distributions directly from raw data, and to remain more balanced than other previous methods. Finally, an application to learn a fuzzy control system, using this approach, will be presented. [ABSTRACT FROM AUTHOR]

This paper presents a clustering approach that integrates multi-objective optimization, weighted k-means and validity analysis in an iterative process to automatically estimate the number of clusters, and then partition the whole given data to produce the most natural clustering. The proposed approach has been tested on real-life dataset; results of both weighted and unweighed k-means are reported to demonstrate applicability and effectiveness of the proposed approach. [ABSTRACT FROM AUTHOR]

Melt index is considered important quality variable determining product specifications. Reliable prediction of melt index (MI) is crucial in quality control of practical propylene polymerization processes. In this paper, a radial basis function network (RBF) model based on principal component analysis (PCA) and genetic algorithm (GA) is developed to infer the MI of polypropylene from other process variables. Considering that the genetic algorithm need long time to converge, chaotic series are explored to get more effective computation rate. The PCA-RBF model is also developed as a basis of comparison research. Brief outlines of the modeling procedure are presented, followed by the procedures for training and validating the model. The research results confirm the effectiveness of the presented methods. [ABSTRACT FROM AUTHOR]

This paper proposes a new face recognition approach by using the Discrete Cosine Transform (DCT) and Hierarchical Radial Basis Function Network (HRBF) classification model. The DCT is employed to extract the input features to build a face recognition system, and the HRBF is used to identify the faces. Based on the pre-defined instruction/operator sets, a HRBF model can be created and evolved. This framework allows input features selection. The HRBF structure is developed using Extended Compact Genetic Programming (ECGP) and the parameters are optimized by Differential Evolution (DE). Empirical results indicate that the proposed framework is efficient for face recognition. [ABSTRACT FROM AUTHOR]

In this paper we propose some methods to build a kernel matrix for classification purposes using Support Vector Machines (SVMs) by fusing polynomial kernels. The proposed techniques have been successfully evaluated on artificial and real data sets. The new methods outperform the best individual kernel under consideration and they can be used as an alternative to the parameter selection problem in polynomial kernel methods. [ABSTRACT FROM AUTHOR]

Hydrological Simulation Program Fortan (HSPF) is a mathematical modelling program developed by the United States Environmental Protection Agency (EPA). HSPF is used for simulating of watershed hidrology and water quality. In this paper, an evolutionary algorithm is applied to automated watershed model calibration. The calibration stage of the model is very important in order to reduce the errors in hydrological predictions. Results show the capacity of the proposed method to simulate watershed streamflows. [ABSTRACT FROM AUTHOR]

This paper studies the structural robustness of evolutionary models of cooperation, i.e. their sensitivity to small structural changes. To do this, we focus on the Prisoner's Dilemma game and on the set of stochastic strategies that are conditioned on the last action of the player's opponent. Strategies such as Tit-For-Tat (TFT) and Always-Defect (ALLD) are particular and classical cases within this framework; here we study their potential appearance and their evolutionary robustness, as well as the impact of small changes in the model parameters on their evolutionary dynamics. Our results show that the type of strategies that are likely to emerge and be sustained in evolutionary contexts is strongly dependent on assumptions that traditionally have been thought to be unimportant or secondary (number of players, mutation-rate, population structure...). We find that ALLD-like strategies tend to be the most successful in most environments, and we also discuss the conditions that favor the appearance of TFTlike strategies and cooperation. Keywords: Evolution of Cooperation; Evolutionary Game Theory; Iterated Prisoner's Dilemma; Tit for Tat; Agent-based Modeling. [ABSTRACT FROM AUTHOR]

In recent years, variational Bayesian learning has been used as an approximation of Bayesian learning. In spite of the computational tractability and good generalization performance in many applications, its statistical properties have yet to be clarified. In this paper, we analyze the statistical property in variational Bayesian learning of Bayesian networks which are widely used in information processing and uncertain artificial intelligence. We derive upper bounds for asymptotic variational stochastic complexities of Bayesian networks. Our result theoretically supports the effectiveness of variational Bayesian learning as an approximation of Bayesian learning. [ABSTRACT FROM AUTHOR]

A normal mixture model, which belongs to singular learning machines, is widely used in statistical pattern recognition. In singular learning machines, the Bayesian learning provides the better generalization performance than the maximum likelihood estimation. However, it needs huge computational cost to realize the Bayesian posterior distribution by the conventional Monte Carlo method. In this paper, we propose that the exchange Monte Carlo method is appropriate for the Bayesian learning in singular learning machines, and experimentally show that it provides better generalization performance in the Bayesian learning of a normal mixture model than the conventional Monte Carlo method. [ABSTRACT FROM AUTHOR]

The self-organizing map (SOM) [1] is used in data analysis for resolving and visualizing nonlinear relationships in complex data. This paper presents an application of the SOM for depicting state and progress of a real-time process. A self-organizing map is used as a visual regression model for estimating the state configuration and progress of an observation in process data. The proposed technique is used for examining full-scope nuclear power plant simulator data. One aim is to depict only the most relevant information of the process so that interpretating process behaviour would become easier for plant operators. In our experiments, the method was able to detect a leakage situation in an early stage and it was possible to observe how the system changed its state as time went on. [ABSTRACT FROM AUTHOR]