Search

The aim of this paper is to discuss the possibility of understanding human social interaction in web communities by analogy with a disease propagation model from epidemiology. When an article is submitted by an individual to a social web service, it is potentially influenced by other participants. The submission sometimes starts a long and argumentative chain of articles, but often does not. This complex behavior makes management of server resources difficult and a more theoretical methodology is required. This paper tries to express these complex human dynamics by analogy with infection by a virus. In this first report, by fitting an epidemiological model to Bulletin Board System (BBS) logs in terms of a numerical triple, we show that the analogy is reasonable and beneficial because the analogy can estimate the community size despite the submitter's information alone being observable. [ABSTRACT FROM AUTHOR]

We develop an agent-based model of the competition between payment cards by focusing on the interactions between consumers and merchants determining the subscription and usage of cards. We find that after a short period of time the market will be dominated by a small number of cards, even though there do not exist significant differences between cards and the market is fully competitive. In contrast to the existing literature we focus on the dynamics of market shares and emergence of multi-homing rather than equilibrium outcomes. [ABSTRACT FROM AUTHOR]

The importance of modularity in product innovation is analyzed in this paper. Through simulations with an agent-based modular economic model, we examine the significance of the use of a modular structure in new product designs in terms of its impacts upon customer satisfaction and firms' competitiveness. To achieve the above purpose, the automatically defined terminal is proposed and is used to modify the simple genetic programming. [ABSTRACT FROM AUTHOR]

In this research, we employ Agent-Based Model to analyze how asset prices are affected by investors' Behavior. This analysis places focus on the influence of overconfident investors on financial market. As a result of intensive analysis, we find that overconfident investors are generated in a bottom-up fashion in the market. Furthermore, it has also been found that overconfident investors have the ability to contribute to market efficiency. [ABSTRACT FROM AUTHOR]

This paper presents a novel hybrid algorithm for combinatorial optimization problems based on mixing the cross-entropy (CE) method and a Hopfield neural network. The algorithm uses the CE method as a global search procedure, whereas the Hopfield network is used to solve the constraints associated to the problems. We have shown the validity of our approach in several instance of the generalized frequency assignment problem. [ABSTRACT FROM AUTHOR]

This paper presents a novel approach to solve multiclass classification problems using pure evolutionary techniques. The proposed approach is called Evolution of Geometric Structures algorithm, and consists in the evolution of several geometric structures such as hypercubes, hyperspheres, hyperoctahedrons, etc. to obtain a first division of the samples space, which will be re-evolved in a second step in order to solve samples belonging to two or more structures. We have applied the EGS algorithm to a well known multiclass classification problem, where our approach will be compared with several existing classification algorithms. [ABSTRACT FROM AUTHOR]

The edited k-nearest neighbor consists of the application of the k-nearest neighbor classifier with an edited training set, in order to reduce the classification error rate. This edited training set is a subset of the complete training set in which some of the training patterns are excluded. In recent works, genetic algorithms have been successfully applied to generate edited sets. In this paper we propose three improvements of the edited k-nearest neighbor design using genetic algorithms: the use of a mean square error based objective function, the implementation of a clustered crossover, and a fast smart mutation scheme. Results achieved using the breast cancer database and the diabetes database from the UCI machine learning benchmark repository demonstrate the improvement achieved by the joint use of these three proposals. [ABSTRACT FROM AUTHOR]

This paper proposes a diversity generating mechanism for an evolutionary algorithm that determines the basic structure of Multilayer Perceptron (MLP) classifiers and simultaneously estimates the coefficients of the models. We apply a modified version of a recently proposed diversity enhancement mechanism [1], that uses a variable population size and periodic partial reinitializations of the population in the form of a saw-tooth function. Our improvement on this standard scheme consists of guiding saw-tooth reinitializations by considering the variance of the best individuals in the population, performing the population restart when the difference of variance between two consecutive generations is lower than a percentage of the previous variance. The empirical results over six benchmark datasets show that the proposed mechanism outperforms the standard saw-tooth algorithm. Moreover, results are very promising in terms of classification accuracy, yielding a state-of-the-art performance. [ABSTRACT FROM AUTHOR]

This paper explores quantitative evaluation criteria for service and potentials of new service from the transportation viewpoint. For this purpose, we analyze transport networks of railway, subway, and waterbus, and have revealed the following implications: (1) efficiency criterion proposed by Latora [7,8] and centrality criterion in the complex network literature can be applied as quantitative evaluation criteria for service in a transportation domain; and (2) new services are highly embedded among networks, i.e., the analyses of the combined networks have the great potential for finding new services that cannot be found by analyzing a single network. [ABSTRACT FROM AUTHOR]

Recommender System provides certain products adapted to a target user, from a large number of products. One of the most successful recommendation algorithms is Collaborative Filtering, and it is used in many websites. However, the recommendation result is influenced by community characteristics such as the number of users and bias of users' preference, because the system uses ratings of products by the users at the recommendation. In this paper, we evaluate an effect of community characteristics on recommender system, using multi-agent based simulation. The results show that a certain number of ratings are necessary to effective recommendation based on collaborative filtering. Moreover, the results also indicate that the number of necessary ratings for recommendation depends on the number of users and bias of the users' preference. [ABSTRACT FROM AUTHOR]

Conjoint analysis is a statistical technique to reveal customers' invisible preference using series of questions regarding tradeoffs in products. In this paper, we propose a new variant of this technique that uses products layout and customers' actions in a store instead of conjoint cards and answers. We demonstrate the effectiveness of this method by making agent-based in-store simulator that can reproduce the congestion in a store. The parameters of the agents in the simulator were determined by our technique - video-based conjoint analysis. [ABSTRACT FROM AUTHOR]

This paper presents the computational results obtained in the strategy experiments in an artificial futures market with human subjects. Participants submit their own strategy files and they receive the performances of all the market participants in order to improve for the next round. After two-round experiments, simulations with only machine agents are run. We find that the time series data support so-called stylized facts in some regards and that experiments of human subjects seem to make the prices be closer to a theoretical value. [ABSTRACT FROM AUTHOR]

This paper illustrates how to compare different agent-based models and how to compare an agent-based model with real data. As examples we investigate ARFIMA models, the probability density function, and the spectral density function. We illustrate the methodology in an analysis of the agent-based model developed by Levy, Levy, Solomon (2000), and confront it with the S&P 500 for a comparison with real life data. [ABSTRACT FROM AUTHOR]

In the past few years, various variants of the self-organising map (SOM) have been proposed to extend its ability for modelling time-series or temporal sequence. Most of them, however, have little connection to, or are over-simplified, autoregressive (AR) models. In this paper, a new extension termed, self-organising mixture autoregressive (SOMAR) network is proposed to topologically cluster time-series segments into underlying generating AR models. It uses autocorrelation values as the similarity measure between the model and the time-series segments. Such networks can be used for modelling nonstationary time-series. Experiments on predicting artificial time-series (Mackey-Glass) and real-world data (foreign exchange rates) are presented and results show that the proposed SOMAR network is a viable and superior to other SOM-based approaches. [ABSTRACT FROM AUTHOR]

Management of extreme events is required of a special consideration, as well as a sufficiently wide time horizon for solvency evaluation. Whereas their classical adjustment is usually carried out with Extreme Value Theory (EVT)-based distributions (namely, the Generalized Pareto Distribution), Evolutionary Techniques have been tried herein to fit the GPD parameters as an optimisation problem. The comparison between classical and evolutionary techniques highlights the accuracy of the evolutionary process. Data adjusted in this paper come from a Spanish motor liability insurance portfolio. [ABSTRACT FROM AUTHOR]

In this paper we propose to do portfolio management using reinforcement learning (RL) and independent factor model. Factors in independent factor model are mutually independent and exhibit better predictability. RL is applied to each factor to capture temporal dependence and provide investment suggestion on factor. Optimal weights on factors are found by portfolio optimization method subject to the investment suggestions and general portfolio constraints. Experimental results and analysis are given to show that the proposed method has better performance when compare to two alternative portfolio management systems. [ABSTRACT FROM AUTHOR]

In this work we perform an automatic data survey to draw up an optimum portfolio, and to automate the one year forecast of a portfolio's payoff and risk, showing the advantages of using formally grounded models in portfolio management and adopting a strategy that ensures, a high rate of return at a minimum risk. The use of neural networks provides an interesting alternative to the statistical classifier. We can take a decision on the purchase or sale of a given asset, using a neural network to classify the process into three decisions: buy, sell or do nothing. [ABSTRACT FROM AUTHOR]

This paper analyzes discrete time portfolio selection models with Lévy processes. We first implement portfolio models under the hypotheses the vector of log-returns follow or a multivariate Variance Gamma model or a Multivariate Normal Inverse Gaussian model or a Brownian Motion. In particular, we propose an ex-ante and an ex-post empirical comparisons by the point of view of different investors. Thus, we compare portfolio strategies considering different term structure scenarios and different distributional assumptions when unlimited short sales are allowed. [ABSTRACT FROM AUTHOR]

To be successful with certain classification problems or knowledge discovery tasks it is not sufficient to look at the available variables at a single point in time, but their development has to be traced over a period of time. It is shown that patterns and sequences of labeled intervals represent a particularly well suited data format for this purpose. An extension of existing classifiers is proposed that enables them to handle this kind of sequential data. Compared to earlier approaches the expressiveness of the pattern language (using Allen et al.'s interval relationships) is increased, which allows the discovery of many temporal patterns common to real-world applications. [ABSTRACT FROM AUTHOR]

The paper aims to provide for the improvement of software estimation research through a new regression model. The study design of the paper is organized as follows. Evaluation of estimation methods based on historical data sets requires that these data sets be representative for current or future projects. For that reason the data set for software cost estimation model the International Software Benchmarking Standards Group (ISBSG) data set Release 9 is used. The data set records true project values in the real world, and can be used to extract information to predict new projects cost in terms of effort. As estimation method regression models are used. The main contribution of this study is the new cost production function that is used to obtain software cost estimation. The new proposed cost estimation function performance is compared with related work in the literature. In the study same calibration on the production function is made in order to obtain maximum performance. There is some important discussion on how the results can be improved and how they can be applied to other estimation models and datasets. [ABSTRACT FROM AUTHOR]

Clustering is used to generate groupings of data from a large dataset, with the intention of representing the behavior of a system as accurately as possible. In this sense, clustering is applied in this work to extract useful information from the electricity price time series. To be precise, two clustering techniques, K-means and Expectation Maximization, have been utilized for the analysis of the prices curve, demonstrating that the application of these techniques is effective so to split the whole year into different groups of days, according to their prices conduct. Later, this information will be used to predict the price in the short time period. The prices exhibited a remarkable resemblance among days embedded in a same season and can be split into two major kind of clusters: working days and festivities. [ABSTRACT FROM AUTHOR]

The paper deals with power system topology verification, being an important problem in real-time power system modeling. In the paper, the approach with use of multi-agent system is proposed. At the beginning, benefits from utilization of the agent technology are presented. Then, a theoretical background for the power system topology verification is described. Next, multi-agent system for topology verification and functions of particular agents are characterized. At the end, features of the presented approach to power system topology verification are summed up. [ABSTRACT FROM AUTHOR]

The present work continues a longer research in the field of flexible transportation services and the design of an agent system devoted to the planning, scheduling and control of trips under such a domain. In particular, this paper focuses in the design and development of the interface agents present in the system by following an agent development methodology named PASSI. The interface agent devoted to interaction with the customers is explained in detail and its prototype is shown. [ABSTRACT FROM AUTHOR]

In multi-agent systems, complex and dynamic interactions often emerge among individual agents. The ability of each agent to learn adaptively is therefore important for them to survive in such changing environment. In this paper, we consider the effects of neighbourhood structure on the evolution of cooperative behaviour in the N-Player Iterated Prisoner's Dilemma (NIPD). We simulate the NIPD as a bidding game on a two dimensional grid-world, where each agent has to bid against its neighbours based on a chosen game strategy. We conduct experiments with three different types of neighbourhood structures, namely the triangular neighbourhood structure, the rectangular neighbourhood structure and the random pairing structure. Our results show that cooperation does emerge under the triangular neighbourhood structure, but defection prevails under the rectangular neighbourhood structure as well as the random pairing structure. [ABSTRACT FROM AUTHOR]

In this paper a new type of interface agent will be presented. This agent is oriented to model systems for human based computation. This kind of computation, that we consider a logical extension of intelligent agent paradigm, emerges as valid approach for the resolution of complex problems. Firstly an study of the state of the art of interface agents will be review. Next, human based computation will be defined and we will see how is necessary to extend the current typology of interface agents to model this new kind of computation. In addition, a new type of interface agent, oriented to model this type of computational system, will be presented. Finally, two of the most representative applications of human based computation will be specified using this new typology. [ABSTRACT FROM AUTHOR]

The power of social values that helps to shape or formulate our behavior patterns is not only inevitable, but also how we have surreptitiously responded to the hidden curriculum that derives from such social values in our decision making can be just as significant. Through a machine learning approach, we are able to discover the agent dynamics that drives the evolution of the social groups in a community. By doing so, we set up the problem by introducing an agent-based hidden Markov model, in which the acts of an agent are determined by micro-laws with unknown parameters. To solve the problem, we develop a multistage learning process for determining the micro-laws of a community based on observed set of communications between actors without the semantic contents. We present the results of extensive experiments on synthetic data as well as some results on real communities, e.g., Enron email and movie newsgroups. [ABSTRACT FROM AUTHOR]

False sharing is a result of co-location of unrelated data in the same unit of memory coherency, and is one source of unnecessary overhead being of no help to keep the memory coherency in multiprocessor systems. Moreover, the damage caused by false sharing becomes large in proportion to the granularity of memory coherency. To reduce false sharing in page-based DSM systems, it is necessary to allocate unrelated data objects that have different access patterns into the separate shared pages. In this paper we propose sized and call-site tracing-based shared memory allocator, shortly SCSTallocator. SCSTallocator expects that the data objects requested from the different call-sites may have different access patterns in the future. So SCSTallocator places each data object requested from the different call-sites into the separate shared pages, and consequently data objects that have the same call-site are likely to get together into the same shared pages. At the same time SCSTallocator places each data object that has different size into different shared pages to prohibit the different-sized objects from being allocated to the same shared page. We use execution-driven simulation of real parallel applications to evaluate the effectiveness of our SCSTallocator. Our observations show that our SCSTallocator outperforms the existing dynamic shared memory allocators. By combining the two existing allocation technique, we can reduce a considerable amount of false sharing misses. [ABSTRACT FROM AUTHOR]

In this paper, a hybrid social model of econometric model and social influence model is proposed to settle the problem in power resources management. And, a hybrid society simulation platform based on the proposed model, termed Residential Electric Power Consumption Multi-Agent Systems (RECMAS), is designed to simulate residential power consumption by multi-agent. RECMAS is composed of consumer agent, power supplier agent, and policy maker agent. It provides the policy makers with an additional tool to evaluate power price policies and public education campaigns in different scenarios. Through an influenced diffusion mechanism, RECMAS can simulate the factors affecting power consumption, and the ones associated with public education campaigns. The application of the method for simulating residential power consumption in China is presented. [ABSTRACT FROM AUTHOR]

This work tries to go a step further in the development of methods based on automatic learning techniques to parse and interpret data relating to cognitive decline (CD). There have been studied the neuropsychological tests of 267 consultations made over 30 patients by the Alzheimer's Patient Association of Gran Canaria in 2005. The Sanger neural network adaptation for missing values treatment has allowed making a Principal Components Analysis (PCA) on the successfully obtained data. The results show that the first three obtained principal components are able to extract information relating to functional, cognitive and instrumental sintomatology, respectively, from the test. By means of these techniques, it is possible to develop tools that allow physicians to quantify, view and make a better pursuit of the sintomatology associated to the cognitive decline processes, contributing to a better knowledge of these ones. [ABSTRACT FROM AUTHOR]

Much work has been done to identify species-specific proteins in sequenced genomes and hence to determine their function. We assumed that such proteins have specific physico-chemical properties that will discriminate them from proteins in other species. In this paper, we examine the validity of this assumption by comparing proteins and their properties from different bacterial species using Support Vector Machines (SVM). We show that by training on selected protein sequence properties, SVMs can successfully discriminate between proteins of different species. This finding takes us a step closer to inferring the functional characteristics of these proteins. [ABSTRACT FROM AUTHOR]

We apply learning vector quantization to the analysis of tiling microarray data. As an example we consider the classification of C. elegans genomic probes as intronic or exonic. Training is based on the current annotation of the genome. Relevance learning techniques are used to weight and select features according to their importance for the classification. Among other findings, the analysis suggests that correlations between the perfect match intensity of a particular probe and its neighbors are highly relevant for successful exon identification. [ABSTRACT FROM AUTHOR]

DNA sequence is an important determinant of the positioning, stability, and activity of nucleosome, yet the molecular basis of these remains elusive. Positioned nucleosomes are believed to play an important role in transcriptional regulation and for the organization of chromatin in cell nuclei. After completing the genome project of many organisms, sequence mining received considerable and increasing attention. Many works devoted a lot of effort to detect the periodicity in DNA sequences, namely, the DNA segments that wrap the Histone protein. In this paper, we describe and apply a dynamic periodicity detection algorithm to discover periodicity in DNA sequences. Our algorithm is based on suffix tree as the underlying data structure. The proposed approach considers the periodicity of alternative substrings, in addition to considering dynamic window to detect the periodicity of certain instances of substrings. We demonstrate the applicability and effectiveness of the proposed approach by reporting test results on three data sets. [ABSTRACT FROM AUTHOR]

Mass Spectrometry (MS) has been applied to the early detection of ovarian cancer. To date, most of the studies concentrated on the so-called whole-spectrum approach, which treats each point in the spectrum as a separate test, due to its better accuracy than the profiling approach. However, the whole-spectrum approach does not guarantee biologically meaningful results and is difficult for biological interpretation and clinical application. Therefore, to develop an accurate profiling technique for early detection of ovarian cancer is required. This paper proposes a novel profiling method for high-resolution ovarian cancer MS data by integrating the Smoothed Nonlinear Energy Operator (SNEO), correlation-based peak selection and Random Forest classifier. In order to evaluate the performance of this novel method without bias, we employed randomization techniques by dividing the data set into testing set and training set to test the whole procedure for many times over. Test results show that the method can find a parsimonious set of biologically meaningful biomarkers with better accuracy than other methods. [ABSTRACT FROM AUTHOR]

Microarray experiments help researches to construct the structure of gene regulatory networks, i.e., networks representing relationships among different genes. Filter and knowledge extraction processes are necessary in order to handle the huge amount of data produced by microarray technologies. We propose regression trees techniques as a method to identify gene networks. Regression trees are a very useful technique to estimate the numerical values for the target outputs. They are very often more precise than linear regression models because they can adjust different linear regressions to separate areas of the search space. In our approach, we generate a single regression tree for each genes from a set of genes, taking as input the remaining genes, to finally build a graph from all the relationships among output and input genes. In this paper, we will simplify the approach by setting an only seed, the gene ARN1, and building the graph around it. The final model might gives some clues to understand the dynamics, the regulation or the topology of the gene network from one (or several) seeds, since it gathers relevant genes with accurate connections. The performance of our approach is experimentally tested on the yeast Saccharomyces cerevisiae dataset (Rosetta compendium). [ABSTRACT FROM AUTHOR]

Microarray techniques have motivated the develop of different methods to extract useful information from a biological point of view. Biclustering algorithms obtain a set of genes with the same behaviour over a group of experimental conditions from gene expression data. In order to evaluate the quality of a bicluster, it is useful to identify specific tendencies represented by patterns on data. These patterns describe the behaviour of a bicluster obtained previously by an adequate biclustering technique from gene expression data. In this paper a new measure for evaluating biclusters is proposed. This measure captures a special kind of patterns with scaling trends which represents quality patterns. They are not contemplated with the previous evaluating measure accepted in the literature. This work is a first step to investigate methods that search biclusters based on the concept of shift and scale invariance. Experimental results based on the yeast cell cycle and the human B-cell lymphoma datasets are reported. Finally, the performance of the proposed technique is compared with an optimization method based on the Nelder-Mead Simplex search algorithm. [ABSTRACT FROM AUTHOR]

The biclustering techniques have the purpose of finding subsets of genes that show similar activity patterns under a subset of conditions. In this paper we characterize a specific type of pattern, that we have called α-pattern, and present an approach that consists in a new biclustering algorithm specifically designed to find α-patterns, in which the gene expression values evolve across the experimental conditions showing a similar behavior inside a band that ranges from 0 up to a pre-defined threshold called α. The α value guarantees the co-expression among genes. We have tested our method on the Yeast dataset and compared the results to the biclustering algorithms of Cheng & Church (2000) and Aguilar & Divina (2005). Results show that the algorithm finds interesting biclusters, grouping genes with similar behaviors and maintaining a very low mean squared residue. [ABSTRACT FROM AUTHOR]

A novel technique to search for functional modules in a protein-protein interaction network is presented. The network is represented by the adjacency matrix associated with the undirected graph modelling it. The algorithm introduces the concept of quality of a sub-matrix of the adjacency matrix, and applies a greedy search technique for finding local optimal solutions made of dense sub-matrices containing the maximum number of ones. An initial random solution, constituted by a single protein, is evolved to search for a locally optimal solution by adding/removing connected proteins that best contribute to improve the quality function. Experimental evaluations carried out on Saccaromyces Cerevisiae proteins show that the algorithm is able to efficiently isolate groups of biologically meaningful proteins corresponding to the most compact sets of interactions. [ABSTRACT FROM AUTHOR]

Dimensionality reduction can often improve the performance of the k-nearest neighbor classifier (kNN) for high-dimensional data sets, such as microarrays. The effect of the choice of dimensionality reduction method on the predictive performance of kNN for classifying microarray data is an open issue, and four common dimensionality reduction methods, Principal Component Analysis (PCA), Random Projection (RP), Partial Least Squares (PLS) and Information Gain(IG), are compared on eight microarray data sets. It is observed that all dimensionality reduction methods result in more accurate classifiers than what is obtained from using the raw attributes. Furthermore, it is observed that both PCA and PLS reach their best accuracies with fewer components than the other two methods, and that RP needs far more components than the others to outperform kNN on the non-reduced dataset. None of the dimensionality reduction methods can be concluded to generally outperform the others, although PLS is shown to be superior on all four binary classification tasks, but the main conclusion from the study is that the choice of dimensionality reduction method can be of major importance when classifying microarrays using kNN. [ABSTRACT FROM AUTHOR]

Many proteins are composed of two or more subunits, each associated with different polypeptide chains. The number and the arrangement of subunits forming a protein are referred to as quaternary structure. The quaternary structure of a protein is important, since it characterizes the biological function of the protein when it is involved in specific biological processes. Unfortunately, quaternary structures are not trivially deducible from protein amino acid sequences. In this work, we propose a protein quaternary structure classification method exploiting the functional domain composition of proteins. It is based on a nearest neighbor condensation technique in order to reduce both the portion of dataset to be stored and the number of comparisons to carry out. Our approach seems to be promising, in that it guarantees an high classification accuracy, even though it does not require the entire dataset to be analyzed. Indeed, experimental evaluations show that the method here proposed selects a small dataset portion for the classification (of the order of the 6.43%) and that it is very accurate (97.74%). [ABSTRACT FROM AUTHOR]

Classification of micro-array data has been studied extensively but only a small amount of research work has been done on classification of micro-array data involving more than two classes. This paper proposes a learning strategy that deals with building a multi-target classifier and takes advantage from well known data mining techniques. To address the intrinsic difficulty of selecting features in order to promote the classification accuracy, the paper considers the use of a set of binary classifiers each of ones is devoted to predict a single class of the multi-classification problem. These classifiers are similar to local experts whose knowledge (about the features that are most correlated to each class value) is taken into account by the learning strategy for selecting an optimal set of features. Results of the experiments performed on a publicly available dataset demonstrate the feasibility of the proposed approach. [ABSTRACT FROM AUTHOR]

There are lots of validation indexes and techniques to study clustering results. Biclustering algorithms have been applied in Systems Biology, principally in DNA Microarray analysis, for the last years, with great success. Nowadays, there is a big set of biclustering algorithms each one based in different concepts, but there are few intercomparisons that measure their performance. We review and present here some numerical measures, new and evolved from traditional clustering validation techniques, to allow comparisons and validation of biclustering algorithms. [ABSTRACT FROM AUTHOR]

Microarray technology produces large amounts of information to be manipulated by analysis methods, such as biclustering algorithms, to extract new knowledge. All-purpose multivariate data visualization tools are usually not enough for studying microarray experiments. Additionally, clustering tools do not provide means of simultaneous visualization of all the biclusters obtained. We present an interactive tool that integrates traditional visualization techniques with others related to bioinformatics, such as transcription regulatory networks and microarray heatmaps, to provide enhanced understanding of the biclustering results. Our aim is to gain insight about the structure of biological data and the behavior of different biclustering algorithms. [ABSTRACT FROM AUTHOR]

In the information age, online collaboration and social networks are of increasing importance and quickly becoming an integral part of our lifestyle. In business, social networking can be a powerful tool to expand a customer network to which a company can sell products and services, or find new partners / employees in a more trustworthy and targeted manner. Identifying new friends or partners, on social networking websites, is usually done via a keyword search, browsing a directory of topics (e.g. interests, geography, or employer) or a chain of social ties (e.g. links to other friends on a user's profile). However there are limitations to these three approaches. Keyword search typically produces a list of ranked results, where traversing pages of ranked results can be tedious and time consuming to explore. A directory of groups / networks is generally created manually, requires significant ongoing maintenance and cannot keep up with rapid changes. Social chains require the initial users to specify metadata in their profile settings and again may no be up to date. In this paper we propose to use the topological tree method to dynamically identify similar groups based on metadata and content. The topological tree method is used to automatically organise social networking groups. The retrieved results, organised using an online version of the topological tree method, are discussed against to the returned results of a social network search. A discussion is made on the criterions of representing social relationships, and the advantages of presenting underlying topics and providing a clear view of the connections between topics. The topological tree has been found to be a superior representation and well suited for organising social networking content. [ABSTRACT FROM AUTHOR]

Life logs include people's experiences collected from various sources. It is used to support user's memory. There are many studies that collect and store life log for personal memory. In this paper, we collect log data from smart phone, derive contexts from the log, and then identify which is meaningful context by using a method based on KeyGraph. To evaluate the proposed method, we show an example of the meaningful places by using contexts and GPS logs collected from two users. [ABSTRACT FROM AUTHOR]

We review a new form of self-organizing map which is based on a nonlinear projection of latent points into data space, identical to that performed in the Generative Topographic Mapping (GTM) [1]. But whereas the GTM is an extension of a mixture of experts, this model is an extension of a product of experts [6]. We show visualisation and clustering results on a data set composed of video data of lips uttering 5 Korean vowels and show that the new mapping achieves better results than the standard Self-Organizing Map. [ABSTRACT FROM AUTHOR]

This work presents a research on the performance capabilities of an extension of the ViSOM (Visualization Induced SOM) algorithm by the use of the ensemble meta-algorithm and a later fusion process. This main fusion process has two different variants, considering two different criteria for the similarity of nodes. These criteria are Euclidean distance and similarity on Voronoi polygons. The capabilities, strengths and weakness of the different variants of the model are discussed and compared more deeply in the present work. The details of several experiments performed over different datasets applying the variants of the fusion to the ViSOM algorithm along with same variants of fusion with the SOM are included for this purpose. [ABSTRACT FROM AUTHOR]

Intrusion Detection Systems (IDS's) monitor the traffic in computer networks for detecting suspect activities. Connectionist techniques can support the development of IDS's by modeling ‘normal' traffic. This paper presents the application of some unsupervised neural methods to a packet dataset for the first time. This work considers three unsupervised neural methods, namely, Vector Quantization (VQ), Self-Organizing Maps (SOM) and Auto-Associative Back-Propagation (AABP) networks. The former paradigm proves quite powerful in supporting the basic space-spanning mechanism to sift normal traffic from anomalous traffic. The SOM attains quite acceptable results in dealing with some anomalies while it fails in dealing with some others. The AABP model effectively drives a nonlinear compression paradigm and eventually yields a compact visualization of the network traffic progression. [ABSTRACT FROM AUTHOR]

Mining association rules is a major technique within data mining and has many applications. Most methods for mining association rules from tabular data mine simple rules which only represent equality in their items. Limiting the operator only to "=" results in many interesting frequent patterns that may exist not being identified. It is obvious that where there is an order between objects, greater than or less than a value is as important as equality. This motivates extension, from simple equality, to a more general set of operators. We address the problem of mining general association rules in tabular data where rules can have all operators { ≤ , ≥ , ≠ , = } in their antecedent part. The proposed algorithm, Mining General Rules (MGR), is applicable to datasets with discrete-ordered attributes and on quantitative discretized attributes. The proposed algorithm stores candidate general itemsets in a tree structure in such a way that supports of complex itemsets can be recursively computed from supports of simpler itemsets. The algorithm is shown to have benefits in terms of time complexity, memory management and has great potential for parallelization. [ABSTRACT FROM AUTHOR]

This paper presents a tool for web usage mining. The aim is centered on providing a tool that facilitates the mining process rather than implement elaborated algorithms and techniques. The tool covers different phases of the CRISP-DM methodology as data preparation, data selection, modeling and evaluation. The algorithms used in the modeling phase are those implemented in the Weka project. The tool has been tested in a web site to find access and navigation patterns. [ABSTRACT FROM AUTHOR]

The successful application of machine learning techniques to industrial problems places various demands on the collaborators. The system designers must possess appropriate analytical skills and technical expertise, and the management of the industrial or commercial partner must be sufficiently convinced of the potential benefits that they are prepared to invest in money and equipment. Vitally, the collaboration also requires a significant investment in time from the end-users in order to provide training data from which the system can (hopefully) learn. This poses a problem if the developed Machine Learning system is not sufficiently accurate, as the users and management may view their input as wasted effort, and lose faith with the process. In this paper we investigate techniques for making early predictions of the error rate achievable after further interactions. In particular we show how decomposing the error in different components can lead to useful predictors of achievable accuracy, but that this is dependent on the choice of an appropriate sampling methodology. [ABSTRACT FROM AUTHOR]