Skip to main content
main-content

Über dieses Buch

The book constitutes the refereed proceedings of the 11th International Conference on Adaptive and Natural Computing Algorithms, ICANNGA 2013, held in Lausanne, Switzerland, in April 2013.

The 51 revised full papers presented were carefully reviewed and selected from a total of 91 submissions. The papers are organized in topical sections on neural networks, evolutionary computation, soft computing, bioinformatics and computational biology, advanced computing, and applications.

Inhaltsverzeichnis

Frontmatter

On Appropriate Refractoriness and Weight Increment in Incremental Learning

Neural networks are able to learn more patterns with the incremental learning than with the correlative learning. The incremental learning is a method to compose an associate memory using a chaotic neural network. The capacity of the network is found to increase along with its size which is the number of the neurons in the network and to be larger than the one with correlative learning. In former work, the capacity was over the direct proportion to the network size with suitable pairs of the refractory parameter and the learning parameter. In this paper, the refractory parameter and the learning parameter are investigated through the computer simulations changing these parameters. Through the computer simulations, it turns out that the appropriate parameters lie near the origin with some relation between them.

Toshinori Deguchi, Junya Fukuta, Naohiro Ishii

Vector Generation and Operations in Neural Networks Computations

To make clear the mechanism of the visual movement is important in the visual system. The problem is how to perceive vectors of the optic flow in the network. First, the biological asymmetric network with nonlinearities is analyzed for generating the vector from the point of the network computations. The results are applicable to the V1 and MT model of the neural networks in the cortex. The stimulus with a mixture distribution is applied to evaluate their network processing ability for the movement direction and its velocity, which generate the vector. Second, it is shown that the vector is emphasized in the MT than the V1. The characterized equation is derived in the network computations, which evaluates the vector properties of processing ability of the network. The movement velocity is derived, which is represented in Wiener kernels. The operations of vectors are shown in the divisive normalization network , which will create curl or divergence vectors in the higher neural network as MST area.

Naohiro Ishii, Toshinori Deguchi, Masashi Kawaguchi, Hiroshi Sasaki

Synaptic Scaling Balances Learning in a Spiking Model of Neocortex

Learning in the brain requires complementary mechanisms: potentiation and activity-dependent homeostatic scaling. We introduce synaptic scaling to a biologically-realistic spiking model of neocortex which can learn changes in oscillatory rhythms using STDP, and show that scaling is necessary to balance both positive and negative changes in input from potentiation and atrophy. We discuss some of the issues that arise when considering synaptic scaling in such a model, and show that scaling regulates activity whilst allowing learning to remain unaltered.

Mark Rowan, Samuel Neymotin

Can Two Hidden Layers Make a Difference?

Representations of multivariable Boolean functions by one and two-hidden-layer Heaviside perceptron networks are investigated. Sufficient conditions are given for representations with the numbers of network units depending on the input dimension

d

linearly and polynomially. Functions with such numbers depending on

d

exponentially or having some weights exponentially large are described in terms of properties of their communication matrices. A mathematical formalization of the concept of “highly-varying functions” is proposed. There is given an example of such function which can be represented by a network with two hidden layers with merely

d

units.

Věra Kůrková, Marcello Sanguineti

Time Series Visualization Using Asymmetric Self-Organizing Map

We propose an asymmetric version of the Self-Organizing Map (SOM) capable to properly visualize datasets consisting of time series. The goal is achieved by introducing an asymmetric coefficient making the asymmetric SOM capable to handle time series. The experiments on the U.S. Stock Market Dataset verify and confirm the effectiveness of the proposed asymmetric SOM extension.

Dominik Olszewski, Janusz Kacprzyk, Sławomir Zadrożny

Intelligence Approaches Based Direct Torque Control of Induction Motor

This paper presents a comparative study of two intelligent techniques to replace conventional comparators and selection table of direct torque control for induction machines, namely fuzzy logic and artificial neural network. The comparison with the conventional direct torque control proves that FL-DTC and NN-DTC reduces the electromagnetic torque ripple, stator flux, and stator current. Simulation results prove the effectiveness and the performances proposed strategies.

Moulay Rachid Douiri, Mohamed Cherkaoui

Classifier Ensembles Integration with Self-configuring Genetic Programming Algorithm

Artificial neural networks and symbolic expression based ensembles are used for solving classification problems. Ensemble members and the ensembling method are generated automatically with the self-configuring genetic programming algorithm that does not need preliminary adjusting. Performance of the approach is demonstrated with real world problems. The proposed approach demonstrates results competitive to known techniques.

Maria Semenkina, Eugene Semenkin

A Multi-objective Proposal Based on Firefly Behaviour for Green Scheduling in Grid Systems

Global warming and climate change are threats that the planet is facing nowadays. Green computing has emerged as a challenge to reduce the energy consumption and pollution footprints of computers. Grid Computing could match the principles of Green Computing as it could exploit and efficiently use processors’ computing power. This paper presents a swarm multi-objective optimization algorithm for scheduling experiments (the job execution) on the Grid. Multi-Objective Firefly Algorithm (MO-FA) is inspired by the brightness attraction among fireflies. One of the main contributions of this work is that the increasing firefly brightness is interpreted as an improvement in response time and energy savings. This would fulfill both conflicting objectives of Grid users: execution time and energy consumption. Results show that MO-FA is a reliable method according to its interquartile range and its comparison with the standard and well-known multi-objective algorithm NSGA-II. Moreover, it performs better than actual grid schedulers as the Workload Management System (WMS) and the Deadline Budget Constraint (DBC).

María Arsuaga-Ríos, Miguel A. Vega-Rodríguez

A Framework for Derivative Free Algorithm Hybridization

Column generation is a basic tool for the solution of large-scale mathematical programming problems. We present a class of column generation algorithms in which the columns are generated by derivative free algorithms, like population-based algorithms. This class can be viewed as a framework to define hybridization of free derivative algorithms. This framework has been illustrated in this article using the Simulated Annealing (SA) and Particle Swarm Optimization (PSO) algorithms, combining them with the Nelder-Mead (NM) method. Finally a set of computational experiments has been carried out to illustrate the potential of this framework.

Jose Luis Espinosa-Aranda, Ricardo Garcia-Rodenas, Eusebio Angulo

PSO-Tagger: A New Biologically Inspired Approach to the Part-of-Speech Tagging Problem

In this paper we present an approach to the part-of-speech tagging problem based on particle swarm optimization. The part-of-speech tagging is a key input feature for several other natural language processing tasks, like phrase chunking and named entity recognition. A tagger is a system that should receive a text, made of sentences, and, as output, should return the same text, but with each of its words associated with the correct part-of-speech tag. The task is not straightforward, since a large percentage of words have more than one possible part-of-speech tag, and the right choice is determined by the part-of-speech tags of the surrounding words, which can also have more than one possible tag. In this work we investigate the possibility of using a particle swarm optimization algorithm to solve the part-of-speech tagging problem supported by a set of disambiguation rules. The results we obtained on two different corpora are amongst the best ones published for those corpora.

Ana Paula Silva, Arlindo Silva, Irene Rodrigues

Training Support Vector Machines with an Heterogeneous Particle Swarm Optimizer

Support vector machines are classification algorithms that have been successfully applied to problems in many different areas. Recently, evolutionary algorithms have been used to train support vector machines, which proved particularly useful in some multi-objective formulations and when indefinite kernels are used. In this paper, we propose a new heterogeneous particle swarm optimization algorithm, called scouting predator-prey optimizer, specially adapted for the training of support vector machines. We compare our algorithm with two other evolutionary approaches, using both positive definite and indefinite kernels, on a large set of benchmark problems. The experimental results confirm that the evolutionary algorithms can be competitive with the classic methods and even superior when using indefinite kernels. The scouting predator-prey optimizer can train support vector machines with similar or better classification accuracy than the other evolutionary algorithms, while requiring significantly less computational resources.

Arlindo Silva, Teresa Gonçalves

Fitness Landscape-Based Characterisation of Nature-Inspired Algorithms

A significant challenge in nature-inspired algorithmics is the identification of specific characteristics of problems that make them harder (or easier) to solve using specific methods. The hope is that, by identifying these characteristics, we may more easily predict which algorithms are best-suited to problems sharing certain features. Here, we approach this problem using fitness landscape analysis. Techniques already exist for measuring the “difficulty” of specific landscapes, but these are often designed solely with evolutionary algorithms in mind, and are generally specific to discrete optimisation. In this paper we develop an approach for comparing a wide range of continuous optimisation algorithms. Using a fitness landscape generation technique, we compare six different nature-inspired algorithms and identify which methods perform best on landscapes exhibiting specific features.

Matthew Crossley, Andy Nisbet, Martyn Amos

Evolutionary Generation of Small Oscillating Genetic Networks

We discuss the implementation and results of an evolutionary algorithm designed to generate oscillating biological networks. In our algorithm we have used a type of fitness function which defines oscillations independent of amplitude and period, which improves results significantly when compared to a simple fitness function which only measures the distance to a predefined target function. We show that with our fitness function, we are able to conduct an analysis of minimal oscillating motifs. We find that there are several different examples of mechanisms that generate oscillations, which make use in various ways of transcriptional regulations, complex formation and catalytic degradation.

Matthijs van Dorp, Bruno Lannoo, Enrico Carlon

Using Scout Particles to Improve a Predator-Prey Optimizer

We discuss the use of scout particles, or scouts, to improve the performance of a new heterogeneous particle swarm optimization algorithm, called scouting predator-prey optimizer. Scout particles are proposed as a straightforward way of introducing new exploratory behaviors into the swarm, expending minimal extra resources and without performing global modifications to the algorithm. Scouts are used both as general mechanisms to globally improve the algorithm and also as a simple approach to taylor an algorithm to a problem by embodying specific knowledge. The role of each particle and the performance of the global algorithm is tested over a set of 10 benchmark functions and against two state-of-the-art evolutionary optimizers. The experimental results suggest that, with the addition of scout particles, the new optimizer can be competitive and even superior to the other algorithms, both in terms of performance and robustness.

Arlindo Silva, Ana Neves, Teresa Gonçalves

QR-DCA: A New Rough Data Pre-processing Approach for the Dendritic Cell Algorithm

In this paper, we propose a new approach of data pre- processing based on rough set theory for the Dendritic Cell Algorithm (DCA). Our hybrid immune inspired model, denoted QR-DCA, is based on the functioning of dendritic cells within the framework of rough set theory and more precisely, on the QuickReduct algorithm. As the DCA data pre-processing phase is divided into two sub-steps, feature selection and signal categorization, our QR-DCA model selects the right features for the DCA classification task and categorizes each one of them to its specific signal category. This is achieved while preserving the same DCA main characteristic which is its lightweight in terms of running time. Results show that our new approach generates good classification results. We will also compare our QR-DCA to other rough DCA models to show that our new approach outperforms them in terms of classification accuracy while keeping the worthy characteristics expressed by the DCA.

Zeineb Chelly, Zied Elouedi

Convergence Rates of Evolutionary Algorithms for Quadratic Convex Functions with Rank-Deficient Hessian

The best achievable convergence rates of mutation-based evolutionary algorithms are known for various characteristic test problems. Most results are available for convex quadratic functions with Hessians of full rank. Here, we prove that linear convergence rates are achievable for convex quadratic functions even though the Hessians are rank-deficient. This result has immediate implications for recent convergence results for certain evolutionary algorithms for bi-objective optimization problems.

Günter Rudolph

The Scale-Up Performance of Genetic Algorithms Applied to Group Decision Making Problems

The scale-up performance of genetic algorithms applied to group decision making problems is investigated. Ordinal intervals are used for expressing the individual preferences of the decision makers, as provided independently for each course of action. Genetic algorithms have been found capable of swiftly returning optimal ranking solutions, with computational complexity (the relationship between the number of available courses of action and the number of generations until convergence) expressed by a fourth order polynomial, but found practically independent of the number of decision makers.

Tatiana Tambouratzis, Vassileios Kanellidis

Using Genetic Programming to Estimate Performance of Computational Intelligence Models

This paper deals with the problem of choosing the most suitable model for a new data mining task. The metric is proposed on the data mining tasks space, and similar tasks are identified based on this metric. A function estimating models performance on the new task from both the time and error point of view is evolved by means of genetic programming. The approach is verified on data containing results of several hundred thousands machine learning experiments.

Jakub Šmíd, Roman Neruda

Multi-caste Ant Colony Algorithm for the Dynamic Traveling Salesperson Problem

In this paper we apply a multi-caste ant colony system to the dynamic traveling salesperson problem. Each caste inside the colony contains its own set of parameters, leading to the coexistence of different exploration behaviors. Two multi-caste variants are proposed and analyzed. Results obtained with different dynamic scenarios reveal that the adoption of a multi-caste architecture enhances the robustness of the algorithm. A detailed analysis of the outcomes suggests guidelines to select the best multi-caste variant, given the magnitude and severity of changes occurring in the dynamic environment.

Leonor Melo, Francisco Pereira, Ernesto Costa

Generalized Information-Theoretic Measures for Feature Selection

Information-theoretic measures are frequently employed to select the most relevant subset of features from datasets. This paper focuses on the analysis of continuous-valued features. We compare the common approach with discretization of features prior the analysis, to the direct usage of exact values. Due to the overwhelming costs of computing continuous information-theoretic measures based on Shannon entropy the Renyi and Tsallis generalized measures are considered. To enable computation with continuous Tsallis measures a novel modification of the information potential is introduced. The quality of the analysed measures was assessed indirectly through the classification accuracy in conjuction with the greedy feature selection process. The experiments on datasets from UCI repository show considerable improvements of the results when using both generalized continuous measures.

Davor Sluga, Uros Lotric

PCA Based Oblique Decision Rules Generating

The paper presents the new algorithm of oblique rules induction. On the basis of the initial step that consists in clustering the decision class into subclasses, for every subclass the oblique hypercuboid is generated. Sides of the hypercuboid are parallel and perpendicular to the directions defined by

PCA

. One hypercuboid corresponds to one decision rule. Results of inducting rules in the new way were compared with other oblique and non-oblique rules sets built on the artificial and real data.

Marcin Michalak, Karolina Nurzyńska

Cardinality Problem in Portfolio Selection

There is a variety of models for portfolio selection. However, in portfolio theory applications little or no attention is paid to the cardinality problem. In this paper, an algorithm for dealing with this problem is presented. The proposed allocation algorithm is implemented in a software system, which is based on the Fuzzy Logic Q-measure Model and manages financial investments in real time. Tests on real data from Bulgarian Stock Exchange are presented as illustration to the solution.

Penka Georgieva, Ivan Popchev

Full and Semi-supervised k-Means Clustering Optimised by Class Membership Hesitation

K-Means algorithm is one of the most popular methods for cluster analysis. K-Means, as the majority of clustering methods optimise clusters in an unsupervised way. In this paper we present a method of cluster’s class membership hesitation, which enables k-Means to learn with fully and partially labelled data. In the proposed method the hesitation of cluster during optimisation step is controlled by Metropolis-Hastings algorithm. The proposed method was compared with state-of-art methods for supervised and semi-supervised clustering on benchmark data sets. Obtained results yield the same or better classification accuracy on both types of supervision.

Piotr Płoński, Krzysztof Zaremba

Defining Semantic Meta-hashtags for Twitter Classification

Given the wide spread of social networks, research efforts to retrieve information using tagging from social networks communications have increased. In particular, in Twitter social network,

hashtags

are widely used to define a shared context for events or topics. While this is a common practice often the

hashtags

freely introduced by the user become easily biased. In this paper, we propose to deal with this bias defining semantic meta-hashtags by clustering similar messages to improve the classification. First, we use the user-defined

hashtags

as the Twitter message class labels. Then, we apply the meta-hashtag approach to boost the performance of the message classification.

The meta-hashtag approach is tested in a Twitter-based dataset constructed by requesting public

tweets

to the Twitter API. The experimental results yielded by comparing a baseline model based on user-defined

hashtags

with the clustered meta-hashtag approach show that the overall classification is improved. It is concluded that by incorporating semantics in the meta-hashtag model can have impact in different applications, e.g. recommendation systems, event detection or crowdsourcing.

Joana Costa, Catarina Silva, Mário Antunes, Bernardete Ribeiro

Reinforcement Learning and Genetic Regulatory Network Reconstruction

Many different models of genetic regulatory networks (GRN) exist, but most of them are focused on off-line processing, so that important features of real networks, like adaptive and non-stationary characterare missed. Interdisciplinary insight into the area of self-organization within the living organisms has caused some interesting new thoughts, and the suggested model is among them. Based on reinforcement learning of the Boolean network with random initial structure, the model is searching for a specialized network, that agrees with experimentally obtained data from the real GRN. With some experiments of real biological networks we investigate its behaviour.

Branko Šter, Andrej Dobnikar

Nonlinear Predictive Control Based on Least Squares Support Vector Machines Hammerstein Models

This paper shortly describes nonlinear Model Predictive Control (MPC) algorithms for Least Squares Support Vector Machines (LS-SVM) Hammerstein models. The model consists of a nonlinear steady-state part in series with a linear dynamic part. A linear approximation of the model for the current operating point or a linear approximation of the predicted output trajectory along an input trajectory is used for prediction. As a result, all algorithms require solving on-line a quadratic programming problem or a series of such problems, unreliable and computationally demanding nonlinear optimisation is not necessary.

Maciej Ławryńczuk

Particle Swarm Optimization with Transition Probability for Timetabling Problems

In this paper, we propose a new algorithm to solve university course timetabling problems using a Particle Swarm Optimization (PSO). PSOs are being increasingly applied to obtain near-optimal solutions to many numerical optimization problems. However, it is also being increasingly realized that PSOs do not solve constraint satisfaction problems as well as other meta-heuristics do. In this paper, we introduce transition probability into PSO to settle this problem. Experiments using timetables of the University of Tsukuba showed that this approach is a more effective solution than an Evolution Strategy.

Hitoshi Kanoh, Satoshi Chen

A Consensus Approach for Combining Multiple Classifiers in Cost-Sensitive Bankruptcy Prediction

Bankruptcy prediction is an extremely important topic in the field of financial decision making. There has been a raising interest in studying more accurate predictive models able to provide valuable early warning before the real business failure. Recent researches suggested using the consensus of multiple classifiers for boosting the prediction performance. Yet rarely the cost of misclassification errors is considered in the literature of consensus decision making. In this paper we investigate the performance of classifier ensembles for cost-sensitive bankruptcy prediction. The selection of ensemble members is based on individual performance and pairwise diversity of classifiers. The experimental results on a real world database of French companies show that by selecting appropriate base classifiers the ensemble learning substantially improves the performance of cost-sensitive bankruptcy prediction.

Ning Chen, Bernardete Ribeiro

On the Regularization Parameter Selection for Sparse Code Learning in Electrical Source Separation

Source separation of whole-home electrical consumption also known as energy disaggregation plays a crucial role in energy savings and sustainable development. One important approach towards accurate energy disaggregation is based on sparse code learning. The sparsity-based source separation algorithms allow to build models that explicitly generalize across multiple different devices of the same category. While this method has recently been investigated, yet the importance of the degree of sparseness given by the regularization parameter is rarely considered. In this paper we aim at investigating the performance of learning representations from the aggregated electrical load signal with sparse models for energy disaggregation. In particular we focus our study on the influence of the regularization parameter in the overall approach. The computational experiments yielded in real data from home electrical energy consumption show that for several degrees of sparseness a reliable scheme for energy disaggregation can be obtained with statistical significance.

Marisa Figueiredo, Bernardete Ribeiro, Ana Maria de Almeida

Region Based Fuzzy Background Subtraction Using Choquet Integral

Background subtraction, is a widely used method for identifying moving objects in multimedia applications such as video surveillance. Deterministic approaches are the first applications in literature, and they followed statistical approaches; however, more recently prediction-based filter approaches are preferred by researchers. The methods suggested for background subtraction in traffic surveillance applications, which are subject to many uncertainties, such as illumination noise, sudden changes in ambient light and structural changes, have to date failed to satisfy the requirements. Fuzzy approaches in the Artificial Intelligence method are widely used by researchers to eliminate uncertainties within the problem. In this study, a fuzzy background subtraction method, using choquet integral that process certain group of pixels together in order to eliminate uncertainties is suggested. The method is tested on traffic surveillance dataset, leading to satisfying results.

Muhammet Balcilar, A. Coskun Sonmez

A Robust Fuzzy Adaptive Control Algorithm for a Class of Nonlinear Systems

The paper presents a general methodology of adaptive control based on soft computing models to deal with unknown plants. The problem of parameter estimation is solved using a direct approach, i.e., the controller parameters are adapted without explicitly estimating plant parameters. Thus, very simple adaptive and control laws are constructed within the Lyapunov stability framework. The proposed control ensures global stability of the overall system and convergence of the tracking error to a residual set that depends on the size of unmodelled dynamics. The generality of the approach is substantiated by Stone-Weierstrass theorem, which indicates that any continuous function can be approximated by fuzzy basis function expansion. The hallmarks of the approach are its simplicity and transparency. The paper shows the efficiency of the proposed approach on the control of a heat exchanger.

Sašo Blažič, Igor Škrjanc

Disturbance Measurement Utilization in the Efficient MPC Algorithm with Fuzzy Approximations of Nonlinear Models

Extension of the efficient Model Predictive Control (MPC) algorithm, which uses fuzzy approximations of nonlinear models, with mechanisms of disturbance measurement utilization is proposed. Two methods of disturbance measurement utilization are considered. The first method utilizes a fuzzy model of disturbance influence on the control plant, whereas the second one – a nonlinear model used to obtain the free response. In both methods only the free response generated during the prediction calculation is influenced. Therefore, the prediction has such a form that the MPC algorithm remains to be numerically efficient. Only a quadratic optimization problem must be solved at each iteration in order to derive the control signal. The proposed methods of disturbance measurement utilization can significantly improve control performance offered by the algorithm what is demonstrated in the example control system of a nonlinear chemical CSTR reactor with inverse response.

Piotr M. Marusak

Fast Submanifold Learning with Unsupervised Nearest Neighbors

We present an unsupervised nearest neighbors (UNN) variant for continuous latent spaces that allows to embed patterns in different submanifolds. The problem to simultaneously assign patterns to models and learn the embeddings can be very challenging, as the manifolds may lie closely to each other and can have different dimensions and arbitrary curvature. The UNN-based submanifold learning approach (SL-UNN) that is proposed in this paper combines a fast constructive K-means variant with the UNN manifold learning approach. The resulting speedy approach depends on only few parameters, i.e., a distance threshold to allow the definition of new clusters and the usual UNN parameters. Extensions of SL-UNN are able to automatically determine parameter of each submanifold based on the data space reconstruction error.

Oliver Kramer

Using Carrillo-Lipman Approach to Speed up Simultaneous Alignment and Folding of RNA Sequences

Multiple sequence alignment and RNA folding are two important tasks in the field of bioinformatics. Solving those problems simultaneously leads to biologically more significant results. The only one currently known precise algorithm (Sankoff) is too much computationaly expensive for such long sequences. In this work we introduce a new algorithm, that is a combination of well known Nussinov folding algorithm and Sankoff quadratic alignment algorithm and a speed-up for this algorithm that is inspired by the Carrillo-Lipman algorithm for the multiple sequence alignment problem. This approach may allow us to simultaneously align and fold more than two sequences in a better time than we can do it using the Sankoff algorithm.

Mária Šimaľová

Large Scale Metabolic Characterization Using Flux Balance Analysis and Data Mining

Genome-scale metabolic models of several microbes have been reconstructed from sequenced genomes in the last years. These have been used in several applications in Biotechnology and biological discovery, since they allow to predict the phenotype of the microorganism in distinct environmental or genetic conditions, using for instance Flux Balance Analysis (FBA). This work proposes an analysis workflow using a combination of FBA and Data Mining (DM) classification methods, aiming to characterize the metabolic behaviour of microorganisms using the available models. This framework allows the large scale comparison of the metabolism of different organisms and the prediction of gene expression patterns. Also, it can provide insights about transcriptional regulatory events leading to the predicted metabolic behaviour. DM techniques, namely decision tree and classification rules inference, are used to provide patterns of gene expression based on environmental conditions (presence/ absence of substrates in the media). The methods proposed are applied to the study of the metabolism of two related microbes:

Escherichia coli

and

Salmonella typhimurium

.

Miguel Rocha

Automatic Procedures to Assist in Manual Review of Marine Species Distribution Maps

Ecological Niche Modeling (ENM) is a branch of biology that uses algorithms to predict the distribution of species in a geographic area on the basis of a numerical representation of their preferred habitat and environment. Algorithmic maps can be produced for suitable or native habitats and require a review by human experts. During the review operation biologists use their knowledge about a species to modify the maps. They usually take algorithmic maps as starting point in the review. In this paper we provide a methodology for biologists to use the automatic maps as references also during and after the review process. Our approach is based on a comparison between the reviewed map and two systems: an expert system and a Feed Forward Neural Network. Furthermore we suggest an evaluation procedure of the quality of the environmental features used as training set, for assessing the models reliability.

Gianpaolo Coro, Pasquale Pagano, Anton Ellenbroek

Mining the Viability Profiles of Different Breast Cancer: A Soft Computing Perspective

Cancer cells present several mutations that allow them to grow faster than normal cells, at the time that enables them to avoid apotosis and other control processes. Cancer cell may be affected by synthetic lethality, which refers to the induction of one or more mutations that affect them, but affect normal cells as little as possible. It is one of the goals of bioinformatics to identify synthetic mutations in order to target specific cancers. If synthetic mutations affect several cancer cells, then it is possible that also some normal cells may be affected. In this contribution, we describe a methodology able to identify a small set of those mutations that affect in a differential way several breast cancer lines. Our methodology is an instance of the feature selection problem and based in genetic algorithms for the exploration of the solution space, but guided by mutual information. Our results show that cancer lines can be profiled with only a small subset of mutations from an original list of hundreds of mutations.

Antonio Neme

Image Representation and Processing Using Ternary Quantum Computing

In this paper we address the recently outlined field of Quantum Image Processing and propose a novel model for representing a quantum image. In our approach we use multilevel quantum systems to store and process images because of their advantages in terms of dimension of the available Hilbert space, computational power, physical implementation and security of quantum cryptographic protocols. In particular, we focus on the quantum image representation using qutrits (3-level quantum systems) and discuss possible implementations for basic image processing tasks such as image complement, image binarization and histogram computation.

Simona Caraiman, Vasile Manta

Firefly-Inspired Synchronization of Sensor Networks with Variable Period Lengths

Synchrony is an important requirement in wireless sensor networks. Biologically inspired synchronization has received significant scientific attention for its simplicity and robustness. This paper presents a self-organized approach based on two established synchronization methods, Phase-Advance and the Reachback Firefly Algorithm, that can be used if the phase and the period length of a node must both be adjusted. By considering nodes with different period lengths, we account for and synchronize nodes with variable delays and clock inaccuracies. We evaluate our modifications through extensive simulations with a realistic model.

Stefan Wieser, Pier Luca Montessoro, Mirko Loghi

Phase Transitions in Fermionic Networks

We show that the emergence of different structures in complex networks can be represented in terms of a phase transition for quantum gases. In particular, we propose a model of fermionic networks that allows to investigate the network evolution and its dependence on the system temperature. Simulations, performed in accordance with the cited model, illustrate that the transition from classical random networks to scale-free networks mimics a cooling process in quantum gases. Furthermore, we found that, at very low temperatures, a winner-takes-all structure emerges. We deem this model useful for studying the evolution of complex networks and also for representing competitive dynamics.

Marco Alberto Javarone, Giuliano Armano

New Selection Schemes in a Memetic Algorithm for the Vehicle Routing Problem with Time Windows

This paper presents an extensive study on the pre- and post-selection schemes in a memetic algorithm (MA) for solving the vehicle routing problem with time windows. In the MA, which is a hybridization of the genetic and local optimization algorithms, the population of feasible solutions evolves with time. The fitness of the individuals is measured based on the fleet size and the total distance traveled by the vehicles servicing a set of geographically scattered customers. Choosing the proper selection schemes is crucial to avoid the premature convergence of the search, and to keep the balance between the exploration and exploitation during the search. We propose new selection schemes to handle these issues. We present how the various selection schemes affect the population diversity, convergence of the search and solutions quality. The quality of the solutions is measured as their proximity to the best currently-known feasible solutions. We present the experimental results for the well-known Gehring and Homberger’s benchmark tests.

Jakub Nalepa, Zbigniew J. Czech

Classification Based on the Self-Organization of Child Patients with Developmental Dysphasia

Involvement of mathematical and engineering methods in medicine makes it possible to perform research into processes in the human body by non-invasive methods. Our team cooperates with neurologists in the domain of developmental dysphasia. We search for correlations between the results of EEG, magnetic resonance (MR) tractography, speech signal analysis, clinical speech therapy and psychology. Our aim is to verify a hypothesis of the possibility of classifying and visual representing changes in pathological speech by means of artificial neural networks. This contribution concentrates on one part of this research: disordered children’s speech analysis and results from MR tractography. We try to divide the patients into three groups according to disorder relevance. For classification, we use PCA and SSOM. Evaluation of the results and preparation of a software pack with a user-friendly interface can facilitate the emergence of disease monitoring and improve the quality of therapy.

Jana Tuckova, Josef Vavrina, Jan Sanda, Martin Kyncl

Similarity Analysis Based on Bose-Einstein Divergences for Financial Time Series

Similarity assessment between financial time series is one of problems where the proper methodological choice is very important. The typical correlation approach can lead to misleading results. Often the similarity measure is opposite to the visual observations, expert’s knowledge and even a common sense. The reasons of that can be associated with the properties of the correlation measure and its adequateness for analyzed data, as well as in terms of methodological aspects. In this article, we indicate disadvantages associated with the use of correlation to assess the similarity of financial time series and propose an alternative solution based on divergence measures. In particular, we focus on the Bose-Einstein divergence. The practical experiments conducted on simulated and real data confirmed our concept.

Ryszard Szupiluk, Tomasz Ząbkowski

Exploratory Text Analysis: Data-Driven versus Human Semantic Similarity Judgments

We present an approach for comparing human-made and automatically generated semantic representations with an assumption that neither of these has a primary status over the other. In the experimental part, we compare the results gained by using independent component analysis and the self-organizing map algorithm on word context analysis with a semantically labeled dictionary called BLESS. The data-driven methods are useful in assessing the quality of the hand-created semantic resources and these resources can be used to evaluate the outcome of the automated process. We present a number of specific findings that go beyond typical quantitative evaluations of the results of data-driven methods in which the manually created resources are usually taken as a gold standard.

Tiina Lindh-Knuutila, Timo Honkela

Linear Support Vector Machines for Error Correction in Optical Data Transmission

Reduction of bit error rates in optical transmission systems is an important task that is difficult to achieve. As speeds increase, the difficulty in reducing bit error rates also increases. Channels have differing characteristics, which may change over time, and any error correction employed must be capable of operating at extremely high speeds. In this paper, a linear support vector machine is used to classify large-scale data sets of simulated optical transmission data in order to demonstrate their effectiveness at reducing bit error rates and their adaptability to the specifics of each channel. For the classification, LIBLINEAR is used, which is related to the popular LIBSVM classifier. It is found that it is possible to reduce the error rate on a very noisy channel to about 3 bits in a thousand. This is done by a linear separator that can be built in hardware and can operate at the high speed required of an operationally useful decoder.

Alex Metaxas, Alexei Redyuk, Yi Sun, Alex Shafarenko, Neil Davey, Rod Adams

Windows of Driver Gaze Data: How Early and How Much for Robust Predictions of Driver Intent?

Previous work has demonstrated that distinct gaze patterns precede certain driving manoeuvres [1,2] and that they can be used to build an artificial neural network model which predicts a driver’s intended manoeuvres [3,4]. This study seeks to move closer towards the goal of using gaze data in Advanced Driver Assistance Systems (ADAS) so that they can correctly infer the intentions of the driver from what is implied by the available incoming data. Drivers’ gaze behaviour was measured in a dynamic driving simulator. The amount of gaze data required to make predictions that manoeuvres will occur and the reliablity of these predictions at increasing pre-manoeuvre times were investigated by using various sized windows of gaze data. The relative difficulty of predicting different manoeuvres and the accuracy of the models at different pre-manoeuvre times are discussed.

Firas Lethaus, Rachel M. Harris, Martin R. K. Baumann, Frank Köster, Karsten Lemmer

Particle Swarm Optimization for Auto-localization of Nodes in Wireless Sensor Networks

In this paper, we consider the problem of auto-localization of the nodes of a static Wireless Sensor Network (WSN) where nodes communicate through Ultra Wide Band (UWB) signaling. In particular, we investigate auto-localization of the nodes assuming to know the position of a few initial nodes, denoted as “beacons”. In the considered scenario, we compare the location accuracy obtained with the widely used Two-Stage Maximum-Likelihood algorithm with that achieved with an algorithm based on Particle Swarming Optimization (PSO). Accurate simulation results show that the latter can significantly outperform the former.

Stefania Monica, Gianluigi Ferrari

Effective Rule-Based Multi-label Classification with Learning Classifier Systems

In recent years, multi-label classification has attracted a significant body of research, motivated by real-life applications such as text classification and medical diagnoses. However, rule-based methods, and especially Learning Classifier Systems (LCS), for tackling such problems have only been sparsely studied. This is the motivation behind our current work that introduces a generalized multi-label rule format and uses it as a guide for further adapting the general Michigan-style LCS framework. The resulting LCS algorithm is thoroughly evaluated and found competitive to other state-of-the-art multi-label classification methods.

Miltiadis Allamanis, Fani A. Tzima, Pericles A. Mitkas

Evolutionary Strategies Algorithm Based Approaches for the Linear Dynamic System Identification

In this paper the method of analytical form linear dynamic system identification is considered. The sample of the output measurements and the input control function are the only information that is required. The main problem was reduced to complex global optimization problem. Any solution that delivers an extremum to the criteria is a representation of a model structure and its parameters in the form of real numbers vector. The complexity of the reduced problem and its characteristics lead one to search for some special optimization technique. In the current research extremum seeking is based on a modification of the evolutionary strategies algorithm.

Ivan Ryzhikov, Eugene Semenkin

A Genetic Algorithm Approach for Minimizing the Number of Columnar Runs in a Column Store Table

Column-oriented database systems, usually referred to as column stores, organize data in a column-wise manner. Column-wise data can be compressed efficiently, improving the performance of large read-mostly data repositories such as data warehouses. Many compression algorithms exploit the similarity among the column values, where repeats of the same value form columnar runs. In this paper we present a genetic algorithm for determining an optimal column sorting order which will minimize the number of columnar runs in a column store table and therefore maximize the RLE-based table compression. Experiments show that the algorithm performs consistently well on synthetic table instances as well as realistic datasets, resulting with higher run-reduction efficiency compared to existing heuristic for solving the given problem.

Jane Jovanovski, Maja Siljanoska, Goran Velinov

Shadow Detection in Complex Images Using Neural Networks: Application to Wine Grape Seed Segmentation

Determining the exact point of ripening and harvesting of the grapes is essential for obtaining a wine of quality. Recent methods for determining the ripening of the grapes are based on visual inspection of the seed. These methods have the advantage of being simple and of low-cost, but they are prone to human error, and a large number of samples are required to be analyzed in order to obtain representative information of the reality. Currently, the analysis of the seed is made using images obtained with a digital camera, which have major problems as the existence of shadows and highlights. This paper proposes a segmentation method of grape seed in complex images based on artificial neural networks and color images. The method is robust to imperfections in the images, which permits that this type of analysis is installed in reality.

Felipe Avila, Marco Mora, Claudio Fredes, Paulo Gonzalez

Backmatter

Weitere Informationen