ICNC 2009 Abstracts


Full Papers
Paper Nr: 4
Title:

A COMBINATION OF CONNECTIONIST SYSTEMS AND EVOLUTIONARY COMPUTATION TECHNIQUES TO ACHIEVE THE OPTIMAL DOMAIN FOR STELLAR SPECTRA SIGNAL PROCESSING

Authors:

Diego Ordóñez, Carlos Dafonte, Bernardino Arcay and Minia Manteiga

Abstract: This paper presents part of the work carried out by Coordination Unit 8 of the GAIA project. GAIA is ESA’s spacecraft which is planned to be operative at the start of 2012 and will carry out an a stereoscopic census of the Galaxy. During the present development cycle, synthetic spectra are used to determine the stellar atmospheric parameters, particularly effective temperatures, superficial gravities, metallicities, possible abundances of alpha elements, and individual abundancies of certain chemical elements. We present the results of the application of genetic algorithms to the selection of relevant information from a set of spectra. This information will subsequently feed an artificial neural network that is in charge of extracting the parameters.

Paper Nr: 19
Title:

LEARNING ALGORITHMS WITH NEIGHBORING INPUTS IN SELF-ORGANIZING MAPS FOR IMAGE RESTORATION

Authors:

Michiharu Maeda, Noritaka Shigei and Hiromi Miyajima

Abstract: This paper presents learning algorithms with neighboring inputs in self-organizing maps for image restoration. Novel approaches are described that neighboring pixels as well as a notice pixel are prepared as an input, and a degraded image is restored according to an algorithm of self-organizing maps. The algorithm creates a map containing one unit for each pixel. Utilizing pixel values as input, image inference is conducted by selforganizing maps. An updating function with threshold according to the difference between input value and inferred value is introduced, so as not to respond to noisy input sensitively. The inference of an original image proceeds appropriately since any pixel is influenced by neighboring pixels corresponding to the neighboring setting. Experimental results are presented in order to show that our approaches are effective in quality for image restoration.

Paper Nr: 26
Title:

A NOVEL REGION BASED IMAGE FUSION METHOD USING DWT AND REGION CONSISTENCY RULE

Authors:

Tanish Zaveri and Mukesh Zaveri

Abstract: This paper proposes a novel region based image fusion scheme using discrete wavelet transform and region consistency rule. In the recent literature, region based image fusion methods show better performance than pixel based image fusion methods. The graph based normalized cutset algorithm is used for image segmentation. The region consistency rule is used to select the regions from discrete wavelet transform decomposed source images. The new MMS fusion rule is also proposed to fuse multimodality images. Proposed method is applied on large number of registered images of various categories of multifocus and multimodality images and results are compared using standard reference based and nonreference based image fusion parameters. It has been observed that simulation results of our proposed algorithm are consistent and more information is preserved compared to earlier reported pixel based and region based methods.

Paper Nr: 33
Title:

A MULTI-VALUED NEURON WITH A PERIODIC ACTIVATION FUNCTION

Authors:

Igor Aizenberg

Abstract: In this paper, a new activation function for the multi-valued neuron (MVN) is presented. The MVN is a neuron with complex-valued weights and inputs/output, which are located on the unit circle. Although the MVN has a greater functionality than a sigmoidal or radial basis function neurons, it has a limited capability of learning highly nonlinear functions. A periodic activation function, which is introduced in this paper, makes it possible to learn nonlinearly separable problems and non-threshold multiple-valued functions using a single multi-valued neuron. The MVN’s functionality becomes higher and the MVN becomes more efficient in solving various classification problems. A learning algorithm based on the error-correction rule for an MVN with the introduced activation function is also presented.

Paper Nr: 43
Title:

A NOVEL DUAL ADAPTIVE NEURO-CONTROLLER BASED ON THE UNSCENTED TRANSFORM FOR MOBILE ROBOTS

Authors:

Marvin K. Bugeja and Simon G. Fabri

Abstract: This paper proposes a novel dual adaptive neuro-control scheme based on the unscented transform for the dynamic control of nonholonomic wheeled mobile robots. The controller is developed in discrete time and the robot nonlinear dynamic functions are unknown to the controller. A multilayer perceptron neural network is used to approximate the nonlinear robot dynamics. The network is trained online via a specifically devised unscented Kalman predictor. In contrast to the majority of adaptive control techniques hitherto proposed in the literature, the controller presented in this paper does not rely on the heuristic certainty equivalence assumption, but accounts for the estimates’ uncertainty via the principle of dual adaptive control. Moreover, the novel dual adaptive control law employs the unscented transform to improve on the first-order Taylor approximations inherent in previously published dual adaptive schemes. Realistic simulations, including comparative Monte Carlo tests, are used to illustrate the effectiveness of the proposed approach.

Paper Nr: 58
Title:

ASSOCIATIVE SELF-ORGANIZING MAP

Authors:

Magnus Johnsson, Christian Balkenius and Germund Hesslow

Abstract: We present a study of a novel variant of the Self-Organizing Map (SOM) called the Associative Self-Organizing Map (A-SOM). The A-SOM is similar to the SOM and thus develops a representation of its input space, but in addition it also learns to associate its activity with the activity of one or several external SOMs. The A-SOM has relevance in e.g. the modelling of expectations in one modality due to the activity invoked in another modality, and in the modelling of the neuroscientific simulation hypothesis. The paper presents the algorithm generalized to an arbitrary number of associated activities together with simulation results to find out about its performance and its ability to generalize to new inputs that it has not been trained on. The simulation results were very encouraging and confirmed the ability of the A-SOM to learn to associate the representations of its input space with the representations of the input spaces developed in two connected SOMs. Good generalization ability was also demonstrated.

Paper Nr: 66
Title:

COMPARISON OF ANFIS AND ORDINARY KRIGING TO ASSESS HYDRAULIC HEAD DISTRIBUTION - The Orgeval Case Study

Authors:

Bedri Kurtulus, Nicolas Flipo, Patrick Goblet, Guillaume Vilain, Julien Tournebize and Gaëlle Tallec

Abstract: In this study, two methods are evaluated for assessing hydraulic head distribution in an aquifer unit. These methods consist in Ordinary Kriging (OK) and Adaptive Neuro Fuzzy based Inference System (ANFIS). Both methods are applied on the same case study: a part of the agricultural basin of the Orgeval located 70 km east of Paris, France. 68 samples were used to predict hydraulic head distribution on a 100 m square - grid. Cartesian coordinates of the samples were used as inputs of the ANFIS, which gives encouraging result. Both simulations have realistic pattern (R2 > 0.97) even if OK performs slightly better than ANFIS at sampling site. Simulated hydraulic head distributions present discrepancies because the two methods capture different patterns. Combined use of the two approaches allow for improving the sampling location of the observation network.

Paper Nr: 68
Title:

A NEURAL NETWORK MODEL OF THE OLFACTORY SYSTEM FOR GLOMERULAR ACTIVITY PREDICTION

Authors:

Zu Soh, Ryuji Inazawa, Toshio Tsuji, Noboru Takiguchi and Hisao Ohtake

Abstract: Recently, the importance of odors has begun to be emphasized as well as methods for their evaluation, especially in the fragrance and food industries. Although odors can be characterized by their odorant components, their chemical information cannot be directly related to the flavors we perceive. Recent research has revealed that neuronal activity related to glomeruli (which form part of the olfactory system) is closely connected to odor qualities. In this paper, we propose a neural network model of the olfactory system in mice to predict glomerular activity from odorant molecules. To adjust the parameters included in the model, a learning algorithm is also proposed. The results of simulation proved that the relationship between glomerular activity and odorant molecules could be approximated using the proposed model. In addition, the model could predict glomerular activity to a certain extent. These results suggest that the proposed model could be utilized to predict odor qualities for future application.

Paper Nr: 80
Title:

FORECASTING WITH NEUROSOLVER

Authors:

Andrzej Bieszczad

Abstract: Neurosolver is a neuromorphic planner and a problem solving system. It was tested on several problem solving and planning tasks such as re-arranging blocks and controlling a software-simulated artificial rat running in a maze. In these tasks, the Neurosolver created models of the problem as temporal patterns in the problem space. These behavioral traces were then used to perform search and generate actions. While exploring general problem capabilities of the Neurosolver, it was observed that the traces of the past in the problem space can also be used for predicting future behavioral patterns. In this paper, we present an analysis of these capabilities in context of the sample data sets made available for the NN5 competition.

Short Papers
Paper Nr: 7
Title:

AUTOMATIC PARALLELIZATION IN NEURAL COMPUTERS

Authors:

João Pedro Neto

Abstract: Neural Networks are more than just mathematical tools to achieve optimization and learning via sub-symbolic computations. Neural networks can perform several other types of computation, namely symbolic and chaotic computations. The discrete time neural model presented here can perform those three types of computations in a modular way. This paper focuses on how neural networks within this model can be used to automatically parallelize computational processes.

Paper Nr: 11
Title:

ON ADAPTIVE MODELING OF NONLINEAR EPISODIC REGIONS IN KSE-100 INDEX RETURNS

Authors:

Rosheena Siddiqi and Syed Nasir Danial

Abstract: This paper employs the Hinich portmanteau bicorrelation test with the windowed testing method to identify nonlinear behavior in the rate of returns series for Karachi Stock Exchange indices. The stock returns series can be described to be comprising of few brief phases of highly significant nonlinearity, followed by long phases in which the returns follow a pure noise process. It has been identified that major political and economic events have contributed to the short bursts of nonlinear behavior in the returns series. Finally, these periods of nonlinear behavior are used to predict the behavior of the rest of the regions using a feedforward neural network and dynamic neural network with Bayesian Regularization Learning. The dynamic neural network outperforms the traditional feedforward networks because Bayesian regularization learning method is used to reduce the training epochs. The time-series generating process is found to closely resemble a white noise process with weak dependence on value at lag one.

Paper Nr: 22
Title:

A MODIFIED K-NEAREST NEIGHBOR CLASSIFIER TO DEAL WITH UNBALANCED CLASSES

Authors:

Akram AlSukker, Ahmed Al-Ani and Amir Atiya

Abstract: We present in this paper a simple, yet valuable improvement to the traditional k-Nearest Neighbor (kNN) classifier. It aims at addressing the issue of unbalanced classes by maximizing the class-wise classification accuracy. The proposed classifier also gives the option of favoring a particular class through evaluating a small set of fuzzy rules. When tested on a number of UCI datasets, the proposed algorithm managed to achieve a uniformly good performance.

Paper Nr: 32
Title:

DOES FISHER INFORMATION CONSTRAIN HUMAN MOTOR CONTROL?

Authors:

Cristopher M. Harris

Abstract: Fisher information places a bound on the error (variance) in estimating a parameter. The nervous system, however, often has to estimate the value of a variable on different occasions over a range of parameter values (such as light intensities or motor forces). We explore the optimal way to distribute Fisher information across a range of forces. We consider the simple integral of Fisher information, and the integral of the square root of Fisher information because this functional is independent of re-parameterization of force. We show that the square root functional is optimised by signal-dependent noise in which the standard deviation of force noise is approximately proportional to the mean force up to about 50% maximum force, which is in good agreement with empirical observation. The simple integral does not fit observations. We also note that the usual Cramer-Rao bound is ‘extended’ with signal-dependent noise, but that this may not be exploited by the biological motor system. We conclude that maximising the integral of the square root of Fisher information can capture the signal dependent noise observed in natural point-to-point movements for forces below about 50% of maximum voluntary contraction.

Paper Nr: 41
Title:

A CONNECTIONIST APPROACH TO PART-OF-SPEECH TAGGING

Authors:

F. Zamora-Martínez, M. J. Castro-Bleda, S. España-Boquera, Salvador Tortajada and P. Aibar

Abstract: In this paper, we describe a novel approach to Part-Of-Speech tagging based on neural networks. Multilayer perceptrons are used following corpus-based learning from contextual and lexical information. The Penn Treebank corpus has been used for the training and evaluation of the tagging system. The results show that the connectionist approach is feasible and comparable with other approaches.

Paper Nr: 48
Title:

EVOLVED DUAL WEIGHT NEURAL ARCHITECTURES TO FACILITATE INCREMENTAL LEARNING

Authors:

John A. Bullinaria

Abstract: This paper explores techniques for improving incremental learning performance for generalization tasks. The idea is to generalize well from past input-output mappings that become available in batches over time, without the need to store past batches. Standard connectionist systems have previously been optimized for this problem using an evolutionary computation approach. Here that approach is explored more generally and rigorously, and dual weight architectures are incorporated into the evolutionary neural network approach and shown to result in improved performance over existing incremental learning systems.

Paper Nr: 53
Title:

ARTIFICIAL NEURAL NETWORK MODEL APPLIED TO A PEM FUEL CELL

Authors:

D. S. Falcão, J. C. M. Pires, C. Pinho, A. M. F. R. Pinto and F. G. Martins

Abstract: This study proposes the simulation of PEM fuel cell polarization curves using artificial neural networks (ANN). Fuel cell performance can be affected by numerous parameters, namely, reactants pressure, humidification temperature, stoichiometric flow ratios and fuel cell temperature. In this work, the influence of relative humidity (RH) of the gases, as well as gases and fuel cell temperatures was studied. A feedforward ANN with three layers was applied to predict the influence of those parameters, simulating the voltage of a fuel cell of 25 cm2 area. Different ANN models were tested, varying the number of neurons in the hidden layer (1 to 6). The model performance was evaluated using the Pearson correlation coefficient (R) and the index of agreement of the second order (d2). The results showed that feedforward ANN can be used with success in order to obtain the optimal operating conditions to improve PEM fuel cell performance.

Paper Nr: 57
Title:

BCCI - A BIDIRECTIONAL CORTICAL COMMUNICATION INTERFACE

Authors:

A. Walter, M. Bensch, D. Brugger, W. Rosenstiel, M. Bogdan, N. Birbaumer and A. Gharabaghi

Abstract: Therapeutic methods based on efferent signals from the patients’ brain have been studied extensively in the field of brain-computer interfaces and applied to paralysed and stroke patients. Invasive stimulation is used as a therapeutic tool for patients with Parkinson disease, intractable chronic pain and other neurological diseases. We give a short review of currently used applications for cortical stimulation for stroke patients and braincomputer interfaces for paralyzed patients. We propose a refined approach for stroke rehabilitation as well as the extension of the use of invasive cortical stimulation to ALS patients with an experimental setup inspired by classical conditioning to facilitate the communication with brain-computer interfaces for LIS and CLIS patients. A closed-loop system is described with sophisticated methods for the identification of recording and stimulation sites, feature extraction and adaptation of stimulation algorithms to the patient in order to design a bidirectional cortical communication interface (BCCI).

Paper Nr: 60
Title:

FEEDBACK CONTROL TAMES DISORDER IN ATTRACTOR NEURAL NETWORKS

Authors:

Maria Pietronilla Penna, Anna Montesanto and Eliano Pessa

Abstract: Typical attractor neural networks (ANN) used to model associative memories behave like disordered systems, as the asymptotic state of their dynamics depends in a crucial (and often unpredictable) way on the chosen initial state. In this paper we suggest that this circumstance occurs only when we deal with such ANN as isolated systems. If we introduce a suitable control, coming from the interaction with a reactive external environment, then the disordered nature of ANN dynamics can be reduced, or even disappear. To support this claim we resort to a simple example based on a version of Hopfield autoassociative memory model interacting with an external environment which modifies the network weights as a function of the equilibrium state coming from retrieval dynamics.

Paper Nr: 65
Title:

PARALLEL REWRITING IN NEURAL NETWORKS

Authors:

Ekaterina Komendantskaya

Abstract: Rewriting systems are used in various areas of computer science, and especially in lambda-calculus, higherorder logics and functional programming. We show that the unsupervised learning networks can implement parallel rewriting. We show how this general correspondence can be refined in order to perform parallel term rewriting in neural networks, for any given first-order term. We simulate these neural networks in the MATLAB Neural Network Toolbox and present the complete library of functions written in the MATLAB Neural Network Toolbox.

Paper Nr: 67
Title:

RECURSIVE SELF-ORGANIZING NETWORKS FOR PROCESSING TREE STRUCTURES - Empirical Comparison

Authors:

Pavol Vančo and Igor Farkaš

Abstract: During the last decade, self-organizing neural maps have been extended to more general data structures, such as sequences or trees. To gain insight into how these models learn the tree data, we empirically compare three recursive versions of the self-organizing map – SOMSD, MSOM and RecSOM – using two data sets with the different levels of complexity: binary syntactic trees and ternary trees of linguistic propositions. We evaluate the models in terms of proposed measures focusing on unit’s receptive fields and on model’s capability to distinguish the trees either in terms of separate winners or distributed map output activation vectors. The models learn to topographically organize the data but differ in how they balance the effects of labels and the tree structure in representing the trees. None of the models could successfully distinguish all vertices by assigning them unique winners, and only RecSOM, being computationally the most expensive model regarding the context representation, could unambiguously distinguish all trees in terms of distributed map output activation.

Paper Nr: 73
Title:

DIFFERENCE OF GAUSSIANS TYPE NEURAL IMAGE FILTERING WITH SPIKING NEURONS

Authors:

Sylvain Chevallier and Sonia Dahdouh

Abstract: This contribution describes a bio-inspired image filtering method using spiking neurons. Bio-inspired approaches aim at identifying key properties of biological systems or models and proposing efficient implementations of these properties. The neural image filtering method takes advantage of the temporal integration behavior of spiking neurons. Two experimental validations are conducted to demonstrate the interests of this neural-based method. The first set of experiments compares the noise resistance of a convolutional difference of Gaussians (DOG) filtering method and the neuronal DOG method on a synthetic image. The other experiment explores the edges recovery ability on a natural image. The results show that the neural-based DOG filtering method is more resistant to noise and provides a better edge preservation than classical DOG filtering method.

Paper Nr: 74
Title:

NEURAL NETWORK COMPUTABILITY OF FACE-BASED ATTRACTIVENESS

Authors:

Joshua Chauvin, Marcello Guarini and Christopher Abeare

Abstract: In this work we have explored facial attractiveness as well as sex classification through the application of feed-forward artificial neural network (ANN) models. Data was collected from participants to compile a face database that was later rated by human raters. The neural network analyzed facial images as pixel-data that was converted into vectors. Prediction was carried out by first training the neural network on a number of images (along with their respective attractiveness ratings) and then testing it on new stimuli in order to make generalizations. There was strong intraclass correlation (ICC) and agreement between the neural network outputs and the human raters on facial attractiveness. This project’s success provides novel evidence for the hypothesis that there are objective regularities in facial attractiveness. In addition, there is some indication that the confidence with which sex classification is performed is related to attractiveness. This paper corroborates the work of others that suggests facial attractiveness judgments can be learned by machines.

Paper Nr: 76
Title:

BRAIN CENTERS MODEL AND ITS APPLICATIONS TO EEG ANALYSIS

Authors:

Ivan Gorbunov and Piotr Semenov

Abstract: This paper presents a new approach to EEG analysis and human functional state discrimination. This is Brain Centers Neural Network model (BCNN-model). We declare BCNN-model fundamentals and recent numerical experiments results. These results approve that model has high accuracy in EEG reproduction and human state discrimination. BCNN-model may have applications in functional state identification and brain exploration.

Paper Nr: 78
Title:

ARTIFICIAL NEURAL NETWORKS LEARNING IN ROC SPACE

Authors:

Cristiano Leite Castro and Antônio Padua Braga

Abstract: In order to control the trade-off between sensitivity and specificity of MLP binary classifiers, we extended the Backpropagation algorithm, in batch mode, to incorporate different misclassification costs via separation of the global mean squared error between positive and negative classes. By achieving different solutions in ROC space, our algorithm improved the MLP classifier performance on imbalanced training sets. In our experiments, standard MLP and SVM algorithms were compared to our solution using real world imbalanced applications. The results demonstrated the efficiency of our approach to increase the number of correct positive classifications and improve the balance between sensitivity and specificity.

Paper Nr: 81
Title:

A COMPUTATIONAL STUDY OF THE DIFFUSE NEIGHBOURHOODS IN BIOLOGICAL AND ARTIFICIAL NEURAL NETWORKS

Authors:

P. Fernández López, C. P. Suárez Araujo and P. García Báez

Abstract: This paper presents a computational study on a fundamental aspect concerning with the dynamic of nitric oxide (NO) both in the biological and artificial neural networks, the Diffuse Neighbourhood (DNB). We apply the compartmental model of NO diffusion as formal tool, using a computational neuroscience point of view. The main objective is the analysis of DNB by the observation of the AI-NOD and CDNB variables, defined in this work. We present a study of influences and dependences with respect to associated features to the NO synthesis-diffusion process, and to the environment where it spreads (non-isotropy and non-homogeneity). It is structured into three sets of experiences which cover the quoted aspects: influence of the NO synthesis process, isolated and multiple processes, influence of distance to the element where NO is synthesized, influence of features of the diffusion environment. The developments have been performed in mono and bi-dimensional environments, with endothelial cell features. The importance of this study is providing the needed formalism to quantify the information representation capacity that a type of NO diffusion-based signalling presents and their implications in many other underlying neural mechanisms as neural recruitment, synchronization of computations between neurons and in the brain activity in general.

Paper Nr: 87
Title:

KEY EXCHANGE PROTOCOL USING PERMUTATION PARITY MACHINES

Authors:

Oscar Mauricio Rayes and Karl-Heinz Zimmermann

Abstract: In recent years it was shown that two artificial neural networks can synchronize by mutual learning. This fact can be used in cryptographic applications such as symmetric key exchange protocols. This paper describes the so-called permutation parity machine, an artificial neural network proposed as a binary variant of the tree parity machine. A key agreement mechanism based on neural synchronization of two permutation parity machines will be defined and the security of the key exchange protocol will be discussed.

Paper Nr: 88
Title:

NEURONS OR SYMBOLS - Why Does OR Remain Exclusive?

Authors:

Ekaterina Komendantskaya

Abstract: Neuro-Symbolic Integration is an interdisciplinary area that endeavours to unify neural networks and symbolic logic. The goal is to create a system that combines the advantages of neural networks (adaptive behaviour, robustness, tolerance of noise and probability) and symbolic logic (validity of computations, generality, higherorder reasoning). Several different approaches have been proposed in the past. However, the existing neurosymbolic networks provide only a limited coverage of the techniques used in computational logic. In this paper, we outline the areas of neuro-symbolism where computational logic has been implemented so far, and analyse the problematic areas. We show why certain concepts cannot be implemented using the existing neuro-symbolic networks, and propose four main improvements needed to build neuro-symbolic networks of the future.

Paper Nr: 90
Title:

A BIOLOGICAL NEURAL NETWORK FOR ROBOTIC CONTROL - Towards a Human Neuroprocessor

Authors:

José M. Ferrández, Victor Lorente, Javier Garrigós and Eduardo Fernández

Abstract: The main objective of this work is to analyze the computing capabilities of human neuroblastoma cultured cells and to define stimulation patterns able to modulate the neural activity in response to external stimuli for controlling an autonomous robot. Multielectrode Arrays Setups have been designed for direct culturing neural cells over silicon or glass substrates, providing the capability to stimulate and record simultaneously populations of neural cells. This paper tries to modulate the natural physiologic responses of human neural cells by tetanic stimulation of the culture. If we are able to modify the selective responses of some cells with a external pattern stimuli over different time scales, the neuroblastoma-cultured structure could be trained to process pre-programmed spatio-temporal patterns. We show that the large neuroblastoma networks developed in cultured MEAs are capable of learning: stablishing numerous and dynamic connections, with modifiability induced by external stimuli.

Posters
Paper Nr: 12
Title:

ON CERTAIN GROUP INVARIANT MERCER KERNELS

Authors:

Bernd-Jürgen Falkowski

Abstract: For the construction of support vector machines Mercer Kernels are of considerable importance. Since the conditions of Mercer’s theorem are hard to verify in general, a systematic (constructive) description of Mercer kernels which are invariant under a transitive group action is provided. As an example kernels on Euclidean space invariant under the Euclidean motion group are treated. En passant a minor but confusing error in a seminal paper due to Gangolli is rectified. In addition an interesting relation to radial basis functions is exhibited.

Paper Nr: 14
Title:

COMPARISON BETWEEN SVM AND ANN FOR MODELING THE CEREBRAL AUTOREGULATION BLOOD FLOW SYSTEM

Authors:

Max Chacón, Claudio Araya, Marcela Muñoz and Ronney B. Panerai

Abstract: The performance of SVMs and ANNs as identifiers of time systems is compared with the purpose of analyzing the Cerebral blood flow Autoregulation System, one of the main systems in the field of cerebral hemodynamics. The main variables of this system are Arterial Blood Pressure (ABP) variations and changes in End-tidal pCO2 (EtCO2). In this work we show that models that have ABP and EtCO2 as input, trained with the SVM, are superior to ANN models in terms of the fit of an unknown set, and they are also more adequate for measuring the influence of EtCO2 on Cerebral Blood Flow Velocity.

Paper Nr: 17
Title:

TRAINING FOURIER SERIES NEURAL NETWORKS TO MAP CLOSED CURVES

Authors:

Krzysztof Halawa

Abstract: The paper presents the closed curve mapping method using several Fourier series neural networks having one input and one output only. The proposed method is also excellently fitted for a lossy compression of closed curves. The method does not require a large number of operations and may be used for multi-dimensional curves. Fourier series neural networks are especially well fitted for described purposes.

Paper Nr: 23
Title:

A NOVEL TECHNIQUE FOR IRIS RECOGNITION SYSTEM

Authors:

Kamal Vahdati Bana, Amin Rezaeian Delui and Amir Azizi

Abstract: In this paper we propose a new feature extraction method for iris recognition based on contourlet transform. Contourlet transform captures the intrinsic geometrical structures of iris image. It decomposes the iris image into a set of directional sub-bands with texture details captured in different orientations at various scales so for reducing the feature vector dimensions we use the method for extract only significant bit and information from normalized iris images. In this method we ignore fragile bits. At last, the feature vector is created by using Co-occurrence matrix properties. For analyzing the desired performance of our proposed method, we use the CASIA dataset, which is comprised of 108 classes with 7 images in each class and each class represented a person. And finally we use SVM and KNN classifier for approximating the amount of people identification in our proposed system. Experimental results show that the proposed increase the classification accuracy and also the iris feature vector length is much smaller versus the other methods.

Paper Nr: 38
Title:

COGNITIVE BIASED ACTION SELECTION STRATEGIES FOR SIMULATIONS OF FINANCIAL SYSTEMS

Authors:

Marco Remondino and Nicola Miglietta

Abstract: In agent based simulations, the many entities involved usually deal with an action selection based on the reactive paradigm: they usually feature embedded strategies to be used according to the stimuli coming from the environment or other entities. This can give good results at an aggregate level, but in certain situations (e.g. Game Theory), cognitive agents, embedded with some learning technique, could give a better representation of the real system. The actors involved in real Social Systems have a local vision and usually can only see their own actions or neighbours’ ones (bounded rationality) and sometimes they could be biased towards a particular behaviour, even if not optimal for a certain situation. In the paper, a new method for cognitive action selection is formally introduced, keeping into consideration an individual bias: ego biased learning. It allows the agents to adapt their behaviour according to a payoff coming from the action they performed at time t-1, by converting an action pattern into a synthetic value, updated at each time, but keeping into account their individual preferences towards specific actions.

Paper Nr: 45
Title:

ANALYSIS OF VARIANCE WITH FUNCTIONAL DATA TO DETECT COLOR CHANGES IN GRANITE

Authors:

T. Rivas, J. Taboada, C. Ordóñez and J. M. Matías

Abstract: Analysis of spectral reflectance curves is useful in many application fields. Despite the functional nature of these curves, statistical methods used to date for analysing these curves (classification, analysis of variance, etc.) have tended to be scalar and do not fully take advantage of the information they contain as functional objects. In this article we applied functional analysis of variance to spectral reflectance curves in order to evaluate the impact of protective treatments on granite colour. The use of raw information on the spectrum means that significant changes are detected that might go unnoticed in models that use scalar values to measure colour. The application of the funcional approach enables information to be obtained on changes whose intensity are statistically significant at each point of the spectrum without the need to perform different analyses for each area of the spectrum. Furthermore, the computational load is no greater than for classical multivariate models.

Paper Nr: 51
Title:

INCREMENTAL LEARNING OF CONVOLUTIONAL NEURAL NETWORKS

Authors:

Dušan Medera and Štefan Babinec

Abstract: Convolutional neural networks provide robust feature extraction with ability to learn complex, highdimensional non-linear mappings from collection of examples. To accommodate new, previously unseen data, without the need of retraining the whole network architecture we introduce an algorithm for incremental learning. This algorithm was inspired by AdaBoost algorithm. It utilizes ensemble of modified convolutional neural networks as classifiers by generating multiple hypotheses. Furthermore, with this algorithm we can work with the confidence score of classification, which can play crucial importance in specific real world tasks. This approach was tested on handwritten numbers classification. The classification error achieved by this approach was highly comparable with non-incremental learning.

Paper Nr: 52
Title:

PREDICTION OF GROUND-LEVEL OZONE CONCENTRATIONS THROUGH STATISTICAL MODELS

Authors:

J. C. M. Pires, F. G. Martins, M. C. Pereira and M. C. M. Alvim-Ferraz

Abstract: This study aims to evaluate the performance of three statistical models: (i) multiple linear regression (MLR), (ii) artificial neural network (ANN) and (iii) multi-gene genetic programming (MGP) for predicting the next day hourly average ozone (O3) concentrations. O3 is an important air pollutant that has several negative impacts. Thus, it is important to develop predictive models to prevent the occurrence of air pollution episodes with a time interval enough to take the necessary precautions. The data were collected in an urban site with traffic influences in Oporto Metropolitan Area, Northern Portugal. The air pollutants data (hourly average concentrations of CO, NO, NO2, NOx and O3), the meteorological data (hourly averages of temperature, relative humidity and wind speed) and the day of week were used as inputs for the models. ANN models presented better results in the training step. However, with regards to the aim of this study, MGP presented the best predictions of O3 concentrations (test step). The good performances of the models showed that MGP is a useful tool to public health protection as it can provide more trustful early warnings to the population about O3 concentrations episodes.

Paper Nr: 62
Title:

MODULE COMBINATION BASED ON DECISION TREE IN MIN-MAX MODULAR NETWORK

Authors:

Yue Wang, Bao-Liang Lu and Zhi-Fei Ye

Abstract: TheMin-Max Modular (M3) Network is the convention solution method to large-scale and complex classification problems. We propose a new module combination strategy using a decision tree for the min-max modular network. Compared with min-max module combination method and its component classifier selection algorithms, the decision tree method has lower time complexity in prediction and better generalizing performance. Analysis of parallel subproblem learning and prediction of these different module combination methods of M3 network show that the decision tree method is easy in parallel.

Paper Nr: 71
Title:

A NEURO-FUZZY EMBEDDED SYSTEM FOR INTELLIGENT ENVIRONMENTS

Authors:

Javier Echanobe, Ines del Campo and Guillermo Bosque

Abstract: Intelligent Environments are endowed with a large number of non-intrusive, embeded electronic systems, such as sensors, microprocessors, actuators, etc. These electronic systems must exhibit intelligent abilities in order to learn and adapt from the users’ habits and preferences. In this paper, we propose a Neuro-Fuzzy electronic embedded system to control several ambient parameters of an intelligent environment, such as temperature, illumination, volume of the sound, and aroma. In particular the PWM-ANFIS model has been selected which provides learning/adaptation features and also fuzzy reasoning. The system is implemented on a reconfigurable device (i.e., FPGA) leading to a small, compact and very efficient electronic system.

Paper Nr: 72
Title:

SMART RECOGNITION SYSTEM FOR THE ALPHANUMERIC - Content in Car License Plates

Authors:

A. Akoum, B. Daya and P. Chauvet

Abstract: A license plate recognition system is an automatic system that is able to recognize a license plate number, extracted from an image device. Such system is useful in many fields and places: parking lots, private and public entrances, border control, theft and vandalism control. In our paper we designed such a system. First we separated each digit from the license plate using image processing tools. Then we built a classifier, using a training set based on digits extracted from approximately 350 license plates. Our approach is considered to identify vehicle through recognizing of its license plate using two different types of neural networks: Hopfield and the multi layer perceptron "MLP". A comparative result has shown the ability to recognize the license plate successfully.

Paper Nr: 84
Title:

A SIMPLE NEURAL-NETWORK ALGORITHM FOR CLASSIFICATION OF LIDAR SIGNALS APPLIED TO FOREST-FIRE DETECTION

Authors:

Andrei B. Utkin, Alexander Lavrov and Rui Vilar

Abstract: Detection of smoke plumes using lidar provides many advantages with respect to passive methods of fire surveillance. However, the great sensitivity of the method results in the detection of many spurious signals. Correspondingly, the automatic lidar surveillance must be provided with effective algorithms of separation of the smoke-plume signatures from irrelevant signals. The paper discusses a simple and robust lidar pattern recognition procedure based on the fast extraction of sufficiently pronounced signal peaks and their classification with a perceptron, whose efficiency is enhanced by a fast nonlinear preprocessing. The algorithm is benchmarked against previously developed artificial-intelligence methods of smoke recognition via Relative Operating Characteristic (ROC curve) analysis.