Advances in Artificial Intelligence and Soft Computing

In this paper a sign-based or semiotic formalism is considered. The concept of sign arose in the framework of semiotics. Neurophysiological and psychological researches indicate sign-based structures, which are the basic elements of the world model of a human subject. These elements are formed during his/her activity and communication. In this formalism it was possible to formulate and solve the problem of goal-setting, i.e. generating the goal of behavior.

The detection of synonyms is a challenge that has attracted many contributions for the possible applications in many areas, including Semantic Web and Information Retrieval. An open challenge is to identify synonyms of a term that are appropriate for a specific domain, not just all the synonyms. Moreover, the execution time is critical when handling big data. Therefore, it is needed an algorithm which can perform accurately and fast in detecting domain-appropriate synonyms on-the-fly. This contribution presents SynFinder which uses WordNet and the web of data. Given a term and a domain in input, WordNet is used for the retrieval of all the synonyms of the term. Then, synonyms which do not appear in web pages related to the domain are eliminated. Our experimentation shows a very good accuracy and computation performance of SynFinder, reporting a mean precision of 0.94 and an average execution time lower than 1 s.

Ro-PAAS is a resource which contains the Romanian verbs and their specific argument and adjunct structures. This resource is linked to our dependency Treebank; we have extracted these structures to the trees and each structure has a list of examples. In this paper, we intend to describe the resource and the modalities in which it can be used in other projects. First, we intend to extract rules for building a hybrid, rule and statistic based parser to quickly increase the dimensions of our Treebank. In the second step, the syntactic structures are in relation with the meaning of the predicates; the structures can be used in programs which need a sense disambiguation. In the third step, we intend to start a comparative study on syntactic structures in the old Romanian language pointing out its differences from the contemporary Romanian, and building some tools rule based for processing old Romanian Language.

The paper describes an experiment consisting in the attempt to quantify word-order properties of three Indo-European languages (Czech, English and Farsi). The investigation is driven by the endeavor to find an objective way how to compare natural languages from the point of view of the degree of their word-order freedom. Unlike similar studies which concentrate either on purely linguistic or purely statistical approach, our experiment tries to combine both – the observations are verified against large samples of sentences from available treebanks, and, at the same time, we exploit the ability of our tools to analyze selected important phenomena (as, e.g., the differences of the word order of a main and a subordinate clause) more deeply.The quantitative results of our research are collected from the syntactically annotated treebanks available for all three languages. Thanks to the HamleDT project, it is possible to search all treebanks in a uniform way by means of a universal query tool PML-TQ. This is also a secondary goal of this paper – to demonstrate the research potential provided by language resources which are to a certain extent unified.

This paper deals with the task of sentential paraphrase identification. We work with Russian but our approach can be applied to any other language with rich morphology and free word order. As part of our ParaPhraser.ru project, we construct a paraphrase corpus and then experiment with supervised methods of paraphrase identification. In this paper we focus on the low-level string, lexical and semantic features which unlike complex deep ones do not cause information noise and can serve as a solid basis for the development of an effective paraphrase identification system. Results of the experiments show that the features introduced in this paper improve the paraphrase identification model based solely on the standard low-level features or the optimized matrix metric used for corpus construction.

In this paper, we present experiments with unsupervised dependency parser without using any part-of-speech tags learned from manually annotated data. We use only unsupervised word-classes and therefore propose fully unsupervised approach of sentence structure induction from a raw text. We show that the results are not much worse than the results with supervised part-of-speech tags.

In the information retrieval (IR) research community, it is commonly accepted that independence assumptions in probabilistic IR models are inaccurate. The need for modeling term dependencies has been stressed in the literature. However, little or nothing has been said on the statistical nature of these dependencies. We investigate statistical measures of term-to-query and document term-to-term pairs dependence, using several test collections. We show that document entropy is highly correlated to dependence, but that high ratios of linearly uncorrelated pairs, do not necessarily mean independent pairs. A robust IR model should then consider both dependence and independence phenomena.

Psychologists have long theorized about the effects of birth order on intellectual development and verbal abilities. Several studies within the field of psychology have tried to prove such theories, however no concrete evidence has been found yet. Therefore, in this paper we present an empirical analysis on the pertinence of traditional Author Profiling techniques. Thus, we re-formulate the problem of identifying developed language abilities by firstborns as a classification problem. Particularly we measure the importance of lexical and syntactic features extracted from a set of 129 speech transcriptions, which were gathered from videos of approximately three minutes length each. Obtained results indicate that both bag of words n-grams and bag of part-of-speech n-grams are able to provide useful information for accurately characterize the language properties employed by firstborns and later-borns. Consequently, our performed experiments helped to validate the presence of distinct language abilities among firstborns and later-borns.

Different strategies are currently studied and applied with the objective to facilitate the acceptability and effective use of Ambient Assisted Living (AAL) applications. One of these strategies is the development of speech-based interfaces to facilitate the communication between the system and the user. In addition to the improvement of communication, the voice of the elder can be also used to automatically classify some paralinguistic phenomena associated with specific mental states and assess the quality of the interaction between the system and the target user. This paper presents our initial work in the construction of these classifiers using an existent spoken dialogue corpus. We present the performance obtained in our models using spoken dialogues from young and older users. We also discuss the further work to effectively integrate these models into AAL applications.

Nowadays, ontologies are widely used in different research areas. Ontology mapping or ontology matching aims at finding correspondences among different ontologies The Ontology Alignment Evaluation Initiative (OAEI) is an international coordinated initiative that organizes evaluation of the increasing number of ontology matching systems. This campaign consisted of 6 tracks gathering 8 test cases and different evaluation modalities. The present paper describes a work, which defines measurements (factors) to increase the quality of matching results obtained in any ontology matching approach or system. These filtering measures have been applied to the OAEI benchmark dataset and produced promising results.

A comparative study of speech recognition performance among systems trained with manually labeled corpora and systems trained with semiautomatically labeled corpora is introduced. An automatic labeling system was designed to generate phoneme labels files for all words within the corpus used to train a system of automatic speech recognition. Speech recognition experiments were performed using the same corpus, first training with manually, and later with automatically generated labels. Results show that the recognition performance is better when the training of selected diccionary, is made with automatic label files than when it is made with manual label files. Not only is the automatic labeling of speech corpora faster than manual labeling, but also it is free from the subjectivity inherent in the manual segmentation performed by specialists. The performance achieved in this work is greater than 96 %.

A description logic program (dl-program) consists of a description logic knowledge base (a terminological box and an assertion box) and a set of rules for a logic program. For such description logic programs, instead of providing the fixed-point semantics for dl-programs by the immediate consequence operator, we propose an algorithm based on the paraconsistent relational model that mimics the immediate consequence operator of dl-programs. We also introduce a dl-paraconsistent relation (dl-relation), which is essential for sending information between description logic and logic programs represented in terms of equations containing paraconsistent relations. The first step in our algorithm is to convert rules, which may contain dl-atoms that enable the flow of information between description logic and logic programs, into paraconsistent relation equations that contain paraconsistent relational algebraic operators. The second step is to determine iteratively the fixed-point semantics for dl-programs using these equations. We will also prove the correctness of both steps of the algorithm.

Covering arrays are used in testing deterministic systems where failures occur as a result of interactions among subsystems. The goal is to reveal if any interaction induces a failure in the system. Application areas include software and hardware testing. A binary covering array CA(N;t,k,2) is an $$N \times k$$ array over the alphabet $$\{0,1\}$$ with the property that each set of t columns contains all the $$2^t$$ possible t-tuples of 0’s and 1’s at least once. In this paper we propose a direct method to construct binary covering arrays using an specific interpretation of binomial coefficients: a binomial coefficient with parameters k and r will be interpreted as the set of all the k-tuples from $$\{0,1\}$$ having r ones and $$k-r$$ zeroes. For given values of k and t, the direct method uses an explicit formula in terms of both k and t to provide a covering array CA(N;t,k,2) expressed as the juxtaposition of a set of binomial coefficients; this covering array will be of the minimum size that can be obtained by any juxtaposition of binomial coefficients. In order to derive the formula, a Branch & Bound (B&B) algorithm was first developed; the B&B algorithm provided solutions for small values of k and t that allowed the identification of the general pattern of the solutions. Like others previously reported methods, our direct method finds optimal covering arrays for $$k = t+1$$ and $$k = t+2$$; however, the major achievement is that nine upper bounds were significantly improved by our direct method, plus the fact that the method is able to set an infinite number of new upper bounds for $$t \ge 7$$ given that little work has been done to compute binary covering arrays for general values of k and t.

The $$\mu $$-calculus is an expressive propositional modal logic augmented with least and greatest fixed-points, and encompasses many temporal, program, dynamic and description logics. The model checking problem for the $$\mu $$-calculus is known to be in NP $$\cap $$ Co-NP. In this paper, we study the model checking problem for the $$\mu $$-calculus extended with graded modalities. These constructors allow to express numerical constraints on the occurrence of accessible nodes (worlds) satisfying a certain formula. It is known that the model checking problem for the graded $$\mu $$-calculus with finite models is in EXPTIME. In the current work, we introduce a linear-time model checking algorithm for the graded $$\mu $$-calculus when models are finite unranked trees.

Selection hyper-heuristics are methods that manage the use of different heuristics and recommend one of them that is suitable for the current problem space under exploration. In this paper we describe a hyper-heuristic model for constraint satisfaction that is inspired in the idea of a lifelong learning process that allows the hyper-heuristic to continually improve the quality of its decisions by incorporating information from every instance it solves. The learning takes place in a transparent way because the learning process is executed in parallel in a different thread than the one that deals with the user’s requests. We tested the model on various constraint satisfaction problem instances and obtained promising results, specially when tested on unseen instances from different classes.

Counting models for two Conjunctive Normal Form formulae (2-CFs), known as the #2SAT problem, is a classic #P complete problem. It is known that if the constraint graph of a 2-CF F is acyclic or contains loops and parallel edges, $$\#2SAT(F)$$ can be computed efficiently. In this paper we address the cyclic case different from loops and parallel edges.If the constraint graph G of a 2-CF F is cyclic, T a spanning tree plus loops and parallel edges of G and $$\overline{T}=G\setminus T$$, what we called its cotree, we show that by building a set partition $$\cup T_i$$ of $$\overline{T}$$, where each $$T_i$$ of the partition is formed by the frond edges of the cycles that are chained via other intersected cycles, then a parametric polynomial deterministic procedure for computing #2SAT with time complexity for the worst case of $$O(2^{k} \cdot poly(|E(T)|))$$ can be obtained, where poly is a polynomial function, and k is the cardinality of the largest set in the partition.This method shows that #2SAT is in the class of fixed-parameter tratable (FPT) problems, where the fixed-parameter k in our proposal, depends on the number of edges of a subcotree of a decomposition of the constraint graph (tree+loops+parallel:cotree) associated to the formula.

A performative-based information capacity calculation metric for MultiAgent Interaction Protocols is developed. The metric uses Shannon’s L-channel combinatorial information capacity calculation idea. In the approach, MultiAgent System (MAS) protocol is modeled as a discrete noiseless channel and provides not data level but language specific information capacity calculation for high level performative-based protocols. The amount of information that flow in MAS due to FIPA-ACL performatives used in given communication protocol is measured and figured out. The unit of measurement is bits per performative. Since the metric is a design time metric, it gives the system designer an opportunity to evaluate his/her design without implementing and testing it. In the application of the proposed metric to FIPA contract net protocol and its iterated versions, we observed a correlation between iteration counts and metric values.

In this paper the application of a new method of bio-inspired optimization based on the self-defense mechanism of plants is presented. Through time the planet has gone through changes, so plants have had to adapt to these changes and adopt new techniques to defend from natural predators (herbivores). Several works have shown that plants have mechanisms of self-defense to protect themselves from predators. When the plants detect the presence of invading organisms this triggers a series of chemical reactions that are released to air and attract natural predators of the invading organism [1, 9, 10]. For the development of this algorithm we consider as a main idea the predator prey mathematical model of Lotka and Volterra, where two populations are considered and the objective is to maintain a balance between the two populations.

Traditionally, games and video games have provided a framework for the study of Artificial Intelligence approaches. The main objective of this work is to verify whether the optimization method based on ant colonies can be applied to the development of a competitive agent in the environment of videogames in real time. One of the important characteristics of the presented work is the optimization calculations to void losing the game fluidity. In addition, another aim of the work is to obtain an architecture for the development of educational type videogames to encourage inexperienced users to interact with artificial intelligence algorithms. The game allows the user to experience the changes in the algorithm given certain changes in the parameters and variables used by the Ant Colony Optimization algorithm itself.

Districting is the redrawing of the boundaries of legislative districts for electoral purposes in such a way that the Federal or state requirements, such as contiguity, population equality, and compactness, are fulfilled. The resulting optimization problem involves the former requirement as a hard constraint while the other two are considered as conflicting objective functions. The solution technique used for many years by the Mexican Federal Electoral Institute was an algorithm based on Simulated Annealing. In this article, we present the system proposed for the electoral districting process in the state of Mexico. This system included, a geographic tool to visualize and edit districting plans, and for first time in Mexico, the use of an Artificial Bee Colony based algorithm that automatically creates redistricting plans.

Particle Swarm Optimization (PSO) algorithm is considered as one of the crowd intelligence optimization algorithms. Dynamic optimization problems in which change(s) may happen over the time are harder to manage than static optimization problems. In this paper an algorithm based on PSO and memory for solving dynamic optimization problems has been proposed. The proposed algorithm uses the memory to store the aging best solutions and uses partitioning for preventing convergence in the population. The proposed approach has been tested on moving peaks benchmark (MPB). The MPB is a suitable problem for simulating dynamic optimization problems. The experimental results on the moving peaks benchmark show that the proposed algorithm is superior to several other well-known and state-of-the-art algorithms in dynamic environments.

Machine-Part Cell Formation Problems consists in organizing a plant as a set of cells, each one of them processing machines containing the same type of parts. In recent years, different meta-heuristic have been used to solve this problem. This paper addresses the problem of Machine-Part Cell Formation by using the Migrating Birds Optimization algorithm. The computational experiments show that in most of the benchmark problems the results obtained from the proposed approach are better than those obtained by other methods which are reported in the literature.

The design of manufacturing cells is a manufacturing strategy that involves the creation of an optimal design of production plants, whose main objective is to minimize movements and exchange of material between these cells. Optimal solution of large scale manufacturing cell design problems (MCDPs) are often computationally unfeasible and only heuristic and approximate methods are able to handle such problems. Artificial fish swarm algorithm (AFSA) belongs to the swarm intelligence algorithms, which based on population search, are able to solve complex optimization problems. In this paper we present an AFSA-based approach to solve the MCDP by using the classic Boctor’s mathematical model. The obtained results show that the proposed algorithm produces optimal solutions for all the 50 studied instances.

Neural network morphology in Artificial Neural Networks (ANN) is typically designed depending on specific learning purposes. Biological neural networks, on the contrary, generate their morphology using biochemical markers secreted by each neuron. Specific features such as molecular signalling, electrochemical alphabet and neurite propagation rules are genetically encoded. However, the environment plays also a critical role in network morphology. Neurites are propagated through tissues to reach target neurons, following paths defined by the diffusion of molecular markers. Neurite paths are affected among other phenomena by competence for synaptic resources and volumetric economy.Along this paper we observe some of the mechanisms of biological morphogenesis and their mathematical models. We analyze neurite navigation in short distances using local random propagation rules. Then, using reaction-difussion patterns, the process of molecular signalling and its influence in network morphology is studied. Finally we combine both strategies to generate morphology in ANN’s.

In this paper, a maximum sensibility neural network was implemented in an embedded system with which was performed a control by imitation of a proposed plant. The plant consists of a cooling system and temperature indicator, the learning of the neural network is given by manually adjusting of output values of the indicators and a fan while input signals are obtained by sensors of temperature and presence, the neural network in run mode is able to interpret these data to automatically adjust the output settings and imitate the process with good performance.

We explore different cases of chaotic behavior in a two-neuron Discrete Time Recurrent Neural Network and found, until now, three cases where chaos is present in different forms. We describe how these cases of chaos are produced in a qualitative way.

The present article gives an extension of the real-valued recurrent neural network topology and its Back-Propagation (BP) learning to the complex-valued one. The BP learning is achieved by the use of diagrammatic rules to obtain the adjoint recurrent neural network topology aimed to propagate the output learning error through it so to learn the neural network weights. Then, this BP learning methodology is applied to the Recurrent Complex-Valued Neural Network (RCVNN) BP-learning using two type RCVNN topologies considering two different kinds of activation functions. After that, the second system identification scheme is incorporated in a total direct complex value control scheme of nonlinear oscillatory plants, introducing also an I-term. The total control scheme contained tree RCVNNs. Furthermore, comparative simulation results of one degree of freedom flexible-joint robot model illustrating system identification and control are obtained. The obtained comparative simulation results confirmed the good quality of the proposed control methodology.

The segmentation of nanotube is an important task for Nanotechnology. The performance of segmentation stage determines the accuracy of the measurement process of nanotube when assessing the quality of nanomaterials. In this work we propose two algorithms for segmenting carbon nanotube images. The first one uses a matched filter bank in the preprocessing step and a neural network for segmenting images from Scanning Electron Microscopy. The second algorithm includes the Perona-Malik filter for enhancing the nanotube information. The segmentation phase is composed by the relaxed Otsu’s threshold and an artificial neural network. This algorithm is applied on images from Transmission Electron Microscopy. After the segmentation, for both algorithms, a preprocessing based on mathematical morphology is carried out. The performance of the proposed algorithms is numerically evaluated by using real image databases. Overall accuracy of 92.74 % and 73.99 % were obtained for the first and second algorithm respectively.

A common task in data analysis is to find the appropriate data sample whose properties allow us to infer the parameters of the data population. The most frequently dilemma related to sampling is how to determine the optimal size of the sample. To solve it, there are typical methods based on asymptotical results from the Central Limit Theorem. However, the effectiveness of such methods is bounded by several considerations as the sampling strategy (simple, stratified, cluster-based, etc.), the size of the population or even the dimensionality of the space of the data. In order to avoid such constraints, we propose a method based on a measure of information of the data in terms of Shannon’s Entropy. Our idea is to find the optimal sample of size N whose information is as similar as possible to the information of the population, subject to several constraints. Finding such sample represents a hard optimization problem whose feasible space disallows the use of traditional optimization techniques. To solve it, we resort to Genetic Algorithms. We test our method with synthetic datasets; the results show that our method is suitable. For completeness, we used a dataset from a real problem; the results confirm the effectiveness of our proposal and allow us to visualize different applications.

The multi-objective evolutionary algorithm based on decomposition (MOEA/D) is an aggregation-based algorithm which has became successful for solving multi-objective optimization problems (MOPs). So far, for the continuous domain, the most successful variants of MOEA/D are based on differential evolution (DE) operators. However, no investigations on the application of DE-like operators within MOEA/D exist in the context of combinatorial optimization. This is precisely the focus of the work reported in this paper. More particularly, we study the incorporation of geometric differential evolution (gDE), the discrete generalization of DE, into the MOEA/D framework. We conduct preliminary experiments in order to study the effectiveness of gDE when coupled with MOEA/D. Our results indicate that the proposed approach is able to outperform the standard version of MOEA/D, when solving a combinatorial optimization problem having between two and four objective functions.

A wireless sensor network collects crucial data for decision making in several domains even under extreme deployment conditions. In this scenario, network availability is usually affected by diverse environment variables. The present approach adapts an evolutionary multi-objective technique in order to get network structures that let to perform data routing efficient in energy consumption. The resulting algorithm, MOR4WSN, comes up from a new solution encoding done to the NSGA-II as well as adapting user-preferences handling even if preference context parameters to optimize are contradictory. MOR4WSN allows optimizing data gathering paths, which contributes to increase network longevity. Experimental evaluation shows that network lifecycle is increased when MOR4WSN is used, compared to other routing mechanisms.

In this participation there is sketched and explained mutual intersection between complex networks and evolutionary computation including summarization of our previous results. It is sketched how dynamics of evolutionary algorithm can be converted into a complex network and based on its properties like degree centrality etc. can be improved performance of used evolutionary algorithm. Results presented here are currently numerical demonstration rather than theoretical mathematical proofs. Paper discusses results from differential evolution, self-organizing migrating algorithm, genetic algorithms and artificial bee colony. We open question whether evolutionary algorithms really create complex network structures and whether this knowledge can be successfully used like feedback for control of evolutionary dynamics and its improvement in order to increase the performance of evolutionary algorithms.

In this paper, we present a hybrid evolutionary algorithm with self-adaptive processes to solve a known project scheduling problem. This problem takes into consideration an optimization objective priority for project managers: to maximize the effectiveness of the sets of human resources assigned to the project activities. The hybrid evolutionary algorithm integrates self-adaptive processes with the aim of enhancing the evolutionary search. The behavior of these processes is self-adaptive according to the state of the evolutionary search. The performance of the hybrid evolutionary algorithm is evaluated on six different instance sets and then is compared with that of the best algorithm previously proposed in the literature for the addressed problem. The obtained results show that the hybrid evolutionary algorithm considerably outperforms the previous algorithm.

This work proposes a new framework for implementing control systems for distributed scheduling. The framework E-HIPS (Extended Hybrid Intelligent Process Scheduler) aims to scale processes in production systems as an extension to the framework HIPS, proposed by the authors in previous work. The original proposal presented a methodology and a set of tools that use the theory of agents and the heuristic search technique Genetic Algorithms (GA) for the implementation of computer systems that have the purpose of managing the scheduling of production processes in the industry. This article proposes an extension to the framework HIPS, by substitution of GA on Memetic Algorithms (MA). The article is an analysis of the problem, under the computational viewpoint, a retrospective of the original proposal, and a new description of the framework with these changes. Aiming to evaluate the framework and its extension, an implementation was made of a control application for scheduling flow to a section of a yarn dyeing industry raw materials for clothing. And a comparison of the results with actual production data obtained from the ERP industry where the system was applied.

We describe in this paper a new approach to enhance the bat algorithm using a fuzzy system to dynamically adapt its parameters. The original method is compared with the proposed method and also compared with genetic algorithms, providing a more complete analysis of the effectiveness of the bat algorithm. Simulation results on a set of benchmark mathematical functions show that the fuzzy bat algorithm outperforms the traditional bat algorithm and genetic algorithms.

The aim of the present work is to describe the performance of a fuzzy agent that is implemented in an embedded system to make an on-line tuning of a embedded PID controller. The fuzzy agent inputs are steady-state error, overshooting and settling time, with this input the fuzzy agent is able to automatically adjust the PID parameter in order to have a better performance. The Peltier cells are used to control the temperature of a small chamber.

In this paper, a Fuzzy Chemical Reaction Algorithm (FCRA) is proposed. In order to overcome the problems of the basic Chemical Reaction Algorithm (CRA), we improve the CRA by proposing a FCRA that takes into account the diversity of the population. Comparative experimental results with benchmark functions show that our proposed method performs much better than the original algorithm in problems with many dimensions.

A new fuzzy Bee Colony Optimization (FBCO) algorithm with dynamic adaptation in the alpha and beta parameters using an Interval Type-2 Fuzzy Logic System is presented in this paper. The Bee Colony Optimization meta-heuristic belongs to the class of Nature-Inspired Algorithms. The objective of the work is based on the use of Interval Type-2 Fuzzy Logic to find the best Beta and Alpha parameter values in BCO. We use BCO specifically for tuning membership functions of the fuzzy controller for stability of the trajectories in a mobile robot. We implemented the IAE and MSE metrics as performance metrics of control. We added perturbations in the model with the pulse generator so that the Interval Type-2 Fuzzy Logic System is better analyzed under uncertainty and to verify that the FBCO shows better results than the traditional BCO.

This paper describes the Particle Swarm Optimization of the Fuzzy integrators in Ensembles of ANFIS (adaptive neuro-fuzzy inferences systems) models for the prediction time series. A chaotic system is considered in this work, which is the Mackey-Glass time series, that is generated from a model is in the form of differential equations. This benchmark time series is used to test of performance of the proposed optimization of the fuzzy integrators in ensemble architecture. We used interval type-2 and type-1 fuzzy systems to integrate the output (forecast) of each Ensemble of ANFIS models. Particle Swarm Optimization (PSO) was used for the optimization of membership function parameters of each fuzzy integrator. In the experiments we optimized Gaussian, Generalized Bell and Triangular membership functions parameters for each of the fuzzy integrators. Simulation results show the effectiveness of the proposed approach. Therefore, a comparison was made against another recent work to validate the performance of the proposed model.

Data clustering means to partition the samples in similar clusters; so that each cluster’s samples have maximum similarity with each other and have a maximum distance from the samples of other clusters. Due to the problem of unsupervised clustering selection of a specific algorithm for clustering a set of unknown data is involved in much risk, and we usually fail to find the best option. Because of the complexity of the issue and inefficacy of basic clustering methods, most studies have been directed toward combined clustering methods. We name output partition of a clustering algorithm as a result. Diversity of the results of an ensemble of basic clusterings is one of the most important factors that can affect the quality of the final result. The quality of those results is another factor that affects the quality of the final result. Both factors considered in recent research of combined clustering. We propose a new framework to improve the efficiency of combined clustering that is based on selection of a subset of primary clusters. Selection of a Proper subset has a crucial role in the performance of our method. The selection is done using intelligent methods. The main ideas of the proposed method for selecting a subset of the clusters are to use the clusters that are stable. This process is done by the intelligent search algorithms. To assess the clusters, stability criteria based on mutual information has been used. At last, the selected clusters are going to be aggregated by some consensus functions. Experimental results on several standard datasets show that the proposed method can effectively improve the complete ensemble method.

It is well-known that every classifier method or algorithm, being Multi-Layer Perceptrons, Decisions Trees or the like, are heavily dependent on data. That is to say, their performance varies significantly whether training data is balanced or not, multi-class or binary, or if classes are defined by numeric or symbolic variables. Some unwanted issues arise, for example, classifiers might be over-trained, or they could present bias or variance, all of which lead to poor performance. The classifiers performance can be analyzed by metrics such as specificity, sensitivity, F-Measure, or the area under the ROC curve. Ensembles of Classifiers are proposed as a means to improve classifications tasks. Classical approaches include Boosting, Bagging and Stacking. However, they do not present cooperation among the base classifiers to achieve a superior global performance. For example, it is desirable that individual classifiers are able to communicate each other what tuples are classified correctly and which are not so errors are not duplicated. We propose an Ensemble of Classifiers that relies on a cooperation mechanism to iteratively improve the performance of both, base classifiers and ensemble. Information Fusion is used to reach a decision. The ensemble is implemented as a Multi-Agent System (MAS), programmed on the JADE platform. The base classifiers are taken from WEKA, as well as the calculation of the performance metrics. We prove the ensemble with a real dataset that is unbalanced, multi-class, and high-dimensional, obtained from a psychoacoustics study.

A novel methodology has been suggested to automatically recognize mine in SONAR data. The suggested framework employs a possibilistic ensemble method to classify SONAR instances as mine or mine-like object. The suggested algorithm minimizes an objective function that merges background identification, multi-algorithm fusion criteria and a learning term. The optimization wants to discover backgrounds as solid clusters in subspaces of the high-dimensional feature-space via a possibilistic semi-supervised learning and feature discrimination. The proposed clustering element allocates a degree of typicality to each data instance in order to recognize and decrease the power of noise instances and outliers. After that the approach results in optimal fusion parameters for each background. The trials on artificial datasets and standard SONAR dataset show that our proposed ensemble does better than individual classifiers and unsupervised local fusion.

At present very large volumes of information are being regularly produced in the world. Most of this information is unstructured, lacking the properties usually expected from, for instance, relational databases. One of the more interesting issues in computer science is how, if possible, may we achieve data mining on such unstructured data. Intuitively, its analysis has been attempted by devising schemes to identify patterns and trends through means such as statistical pattern learning. The basic problem of this approach is that the user has to decide, a priori, the model of the patterns and, furthermore, the way in which they are to be found in the data. This is true regardless of the kind of data, be it textual, musical, financial or otherwise. In this paper we explore an alternative paradigm in which raw data is categorized by analyzing a large corpus from which a set of categories and the different instances in each category are determined, resulting in a structured database. Then each of the instances is mapped into a numerical value which preserves the underlying patterns. This is done using a genetic algorithm and a set of multi-layer perceptron networks. Every categorical instance is then replaced by the adequate numerical code. The resulting numerical database may be tackled with the usual clustering algorithms. We hypothesize that any unstructured data set may be approached in this fashion. In this work we exemplify with a textual database and apply our method to characterize texts by different authors and present experimental evidence that the resulting databases yield clustering results which permit authorship identification from raw textual data.

High-utility itemset mining (HUIM) is an important data mining task with wide applications. In this paper, we propose a novel algorithm named EFIM (EFficient high-utility Itemset Mining), which introduces several new ideas to more efficiently discovers high-utility itemsets both in terms of execution time and memory. EFIM relies on two upper-bounds named sub-tree utility and local utility to more effectively prune the search space. It also introduces a novel array-based utility counting technique named Fast Utility Counting to calculate these upper-bounds in linear time and space. Moreover, to reduce the cost of database scans, EFIM proposes efficient database projection and transaction merging techniques. An extensive experimental study on various datasets shows that EFIM is in general two to three orders of magnitude faster and consumes up to eight times less memory than the state-of-art algorithms d$$^2$$HUP, HUI-Miner, HUP-Miner, FHM and UP-Growth+.

Filtering low-quality workers from data sets labeled via crowdsourcing is often necessary due to the presence of low quality workers, who either lack knowledge on corresponding subjects and thus contribute many incorrect labels to the data set, or intentionally label quickly and imprecisely in order to produce more labels in a short time period. We present two new filtering algorithms to remove low-quality workers, called Cluster Filtering (CF) and Dynamic Classification Filtering (DCF). Both methods can use any number of characteristics of workers as attributes for learning. CF separates workers using k-means clustering with 2 centroids, separating the workers into a high-quality cluster and a low-quality cluster. DCF uses a classifier of any kind to perform learning. It builds a model from a set of workers from other crowdsourced data sets and classifies the workers in the data set to filter. In theory, DCF can be trained to remove any proportion of the lowest-quality workers. We compare the performance of DCF with two other filtering algorithms, one by Raykar and Yu (RY), and one by Ipeirotis et al. (IPW). Our results show that CF, the second-best filter, performs modestly but effectively, and that DCF, the best filter, performs much better than RY and IPW on average and on the majority of crowdsourced data sets.

This paper shows the application of an embedded system with a wireless sensor network to collect atmospheric pollutants data obtained from sensors placed into micro-climates; such dataset provides the information required to test classification algorithms, that helps to develop applications to improve air quality in specific areas.