NCTA 2013 Abstracts


Full Papers
Paper Nr: 7
Title:

Growing Surface Structures

Authors:

Hendrik Annuth and Christian-A. Bohn

Abstract: Strictly iterative approaches derived from unsupervised artificial neural network (ANN) methods have been surprisingly efficient for the application of surface reconstruction from scattered 3D points. This comes from the facts, that on the one hand, ANN are able to robustly cluster samples of arbitrary dimension, size, and complexity, and on the second hand, ANN algorithms can easily be adjusted to specific applications by inventing simple local learning rules without loosing the robustness and convergence behavior of the basic ANN approach. In this work, we break up the idea of having just an ``adjustment'' of the basic unsupervised ANN algorithm but intrude on the central learning scheme and explicitly use learned topology within the training process. We demonstrate the performance of the novel concept in the area of surface reconstruction. In this work, we break up the idea of having just an “adjustment” of the basic unsupervised ANN algorithm but intrude on the central learning scheme and explicitly use the learned topology within the training process. We demonstrate the performance of the novel concept in the area of surface reconstruction.

Paper Nr: 16
Title:

Does Ventriloquism Aftereffect Transfer across Sound Frequencies? - A Theoretical Analysis via a Neural Network Model

Authors:

Elisa Magosso, Filippo Cona, Cristiano Cuppini and Mauro Ursino

Abstract: When an auditory stimulus and a visual stimulus are simultaneously presented in spatial disparity, the sound is perceived shifted toward the visual stimulus (ventriloquism effect). After adaptation to a ventriloquism situation, enduring sound shifts are observed in the absence of the visual stimulus (ventriloquism aftereffect). Experimental studies report discordant results as to aftereffect generalization across sound frequencies, varying from aftereffect staying confined to the sound frequency used during the adaptation, to aftereffect transferring across some octaves of frequency. Here, we present a model of visual-auditory interactions, able to simulate the ventriloquism effect and to reproduce – via Hebbian plasticity rules – the ventriloquism aftereffect. The model is suitable to investigate aftereffect generalization as the simulated auditory neurons code both for spatial and spectral properties of the auditory stimuli. The model provides a plausible hypothesis to interpret the discordant results in the literature, showing that different sound intensities may produce different extents of aftereffect generalization. Model mechanisms and hypotheses are discussed in relation to the neurophysiological and psychophysical literature.

Paper Nr: 17
Title:

Magnitude Sensitive Image Compression

Authors:

Enrique Pelayo , David Buldain and Carlos Orrite

Abstract: This paper introduces the Magnitude Sensitive Competitive Learning (MSCL) algorithm as a reliable and effi- cient approach for selective image compression. MSCL is a neural network that has the property of distributing the unit centroids in certain data-distribution zones according to a target magnitude locally calculated for every unit. This feature can be used for image compression to define the block images that will be compressed by Vector Quantization in a later step. As a result, areas of interest receive a lower compression than other parts in the image. Following this approach higher quality in the salient areas of a compressed image is achieved in relation to other methods.

Paper Nr: 20
Title:

Morphological ECG Analysis for Attention Detection

Authors:

Carlos Carreiras, André Lourenço, Helena Aidos, Hugo Plácido da Silva and Ana Fred

Abstract: The electroencephalogram (EEG) signal, acquired on the scalp, has been extensively used to understand cognitive function, and in particular attention. However, this type of signal has several drawbacks in a context of Physiological Computing, being susceptible to noise and requiring the use of impractical head-mounted apparatuses, which impacts normal human-computer interaction. For these reasons, the electrocardiogram (ECG) has been proposed as an alternative source to assess emotion, which is also continuously available, and related with the psychophysiological state of the subject. In this paper we present a study focused on the morphological analysis of the ECG signal acquired from subjects performing a task demanding high levels of attention. The analysis is made using various unsupervised learning techniques, which are validated against evidence found in a previous study by our team, where EEG signals collected for the same task exhibit distinct patterns as the subjects progress in the task.

Paper Nr: 23
Title:

Investigation of Prediction Capabilities using RNN Ensembles

Authors:

Nijolė Maknickienė and Algirdas Maknickas

Abstract: Modern portfolio theory of investment-based financial market forecasting use probability distributions. This investigation used a neural network architecture, which allows to obtain distribution for predictions. Comparison of the two different models - points based prediction and distributions based prediction - opens new investment opportunities. Dependence of forecasting accuracy on the number of EVOLINO recurrent neural networks (RNN) ensemble was obtained for five forecasting points ahead. This study allows to optimize the computational time and resources required for sufficiently accurate prediction.

Paper Nr: 24
Title:

Gene Ontology Analysis of Colorectal Cancer Biomarkers Probed with Affymetrix and Illumina Microarrays

Authors:

Monika Simjanoska, Ana Madevska Bogdanova and Sasho Panov

Abstract: Colorectal cancer is the fourth most common cause of death worldwide. Recently, many microarray experiments have been done to investigate the expression of the genes in the colorectal tissues and thus, to find the answers for its occurrence. Previously, we used experiments obtained from both Illumina and Affymetrix microarray platforms to analyze the gene expression in healthy and carcinogenic tissues. As a result we got specific sets of biomarkers that we used to build an accurate Bayesian diagnostic system. The high degree of classifier's sensitivity and specificity intrigued us to proceed with the research of the significant genes we discovered, the biomarkers. Therefore, in this paper we aim towards biomarkers identification and the functional groups they are associated with, i.e., we performed gene ontology analysis. Investigating the genes that control the colorectal carcinogenic tissue development is of central importance to the verification of the biomarkers' revealing method's validity. Moreover, we showed the importance of their participation in the prior distributions modeling, which is the key part for achieving an accurate Bayesian classification, regardless their strict disease and disorder association.

Paper Nr: 31
Title:

Artificial Curiosity Emerging Human-like Behaviour - A Fundation for Fully Autonomous Cognitive Machines

Authors:

Dominik Maximilián Ramík, Kurosh Madani and Christophe Sabourin

Abstract: This paper is devoted to autonomous cognitive machines by mean of the design of an artificial curiosity based cognitive system for autonomous high-level knowledge acquisition from visual information. Playing a chief role as well in visual attention as in interactive high-level knowledge construction, the artificial curiosity (e.g. perceptual and epistemic curiosities) is realized through combining visual saliency detection and Machine-Learning based approaches. Experimental results validating the deployment of the investigated system have been obtained using a humanoid robot acquiring visually knowledge about its surrounding environment interacting with a human tutor. As show the reported results and experiments, the proposed cognitive system allows the machine to discover autonomously the surrounding world in which it may evolve, to learn new knowledge about it and to describe it using human-like natural utterances.

Paper Nr: 42
Title:

A Computational Model of Grid Cells based on Dendritic Self-organized Learning

Authors:

Jochen Kerdels and Gabriele Peters

Abstract: In this paper we present a new computational model for grid cells. These cells are neurons in the entorhinal cortex of the hippocampal region that encode allocentric spatial information. They possess a peculiar, triangular firing pattern that spans the entire environment with a virtual lattice. We show that such a firing pattern can emerge from a dendritic, self-organized learning process. A key aspect of the proposed model is the hypothesis that the dendritic tree of a grid cell can behave like a sparse self organizing map that tries to cover its input space as best as possible. We argue, that the encoding scheme used by grid cells is possibly not limited to the description of spatial information and may represent a general principle on how complex information is encoded in higher level brain areas like the hippocampal region.

Short Papers
Paper Nr: 2
Title:

Classification of Power Quality Considering Voltage Sags occurred in Feeders

Authors:

Anderson Roges Teixeira Góes, Maria Teresinha Arns Steiner and Pedro José Steiner Neto

Abstract: In this paper we propose a methodology to classify Power Quality for feeders, based on sags and by the use of KDD technique, establishing a quality level printed in labels. To support the methodology, it was applied to feeders on a substation located in Curitiba, Paraná, Brazil, based on attributes such as sag length, duration and frequency (number of occurrences on a given period of time). In the search for feeders quality classification, on the Data Mining stage, the main stage on KDD process, three different techniques were used in a comparatively way for pattern recognition: Artificial Neural Networks, Support Vector Machines an Genetic Algorithms. Those techniques presented acceptable results in classification feeders with no possible classification using a simplified method based on maximum number of sags. Thus, by printing the label with information and Quality level, utilities companies can get better organized for mitigation procedures, by establishing clear targets.

Paper Nr: 4
Title:

Artificial Neural Networks, Multiple Linear Regression and Decision Trees Applied to Labor Justice

Authors:

Genival Pavanelli, Maria Teresinha Arns Steiner, Alessandra Memari Pavanelli and Deise Maria Bertholdi Costa

Abstract: This paper aims to predict the duration of lawsuits for labor users of the justice system. Thus, we intend to provide forecasts of the duration of a labor lawsuit that gives subsidies to establish an agreement between the parties involved in the processes. The proposed methodology consists in applying and comparing three techniques of the Mathematical Programming area, Artificial Neural Networks (ANN), Multiple Linear Regression (MLR) and Decision Trees in order to obtain the best possible performance for the forecast. Therefore, we used the data from the Labor Forum of São José dos Pinhais, Paraná, Brazil, to do the training of various ANNs, the MLR and the Decision Tree. In several simulations, the techniques were used directly and in others, the Principal Component Analysis (PCA) and / or the coding of attributes were performed before their use in order to further improve their performance. Thus, taking up new data (processes) for which it is necessary to predict the duration of the lawsuit, it will be possible to make up conditions to "diagnose" its length preliminarily at its course. The three techniques used were effective, showing results consistent with an acceptable margin of error.

Paper Nr: 9
Title:

Protein Secondary Structure Prediction using an Optimised Bayesian Classification Neural Network

Authors:

Son T. Nguyen and Colin G. Johnson

Abstract: The prediction of protein secondary structure is a topic that has been tackled by many researchers in the field of bioinformatics. In previous work, this problem has been solved by various methods including the use of traditional classification neural networks with the standard error back-propagation training algorithm. Since the traditional neural network may have a poor generalisation, the Bayesian technique has been used to improve the generalisation and the robustness of these networks. This paper describes the use of optimised classification Bayesian neural networks for the prediction of protein secondary structure. The well-known RS126 dataset was used for network training and testing. The experimental results show that the optimised classification Bayesian neural network can reach an accuracy greater than 75%.

Paper Nr: 12
Title:

Chemoinformatics in Drug Design. Artificial Neural Networks for Simultaneous Prediction of Anti-enterococci Activities and Toxicological Profiles

Authors:

Alejandro Speck-Planche and M. N. D. S. Cordeiro

Abstract: Enterococci are dangerous opportunistic pathogens which are responsible of a huge number of nosocomial infections, displaying intrinsic resistance to many antibiotics. The battle against enterococci by using antimicrobial chemotherapies will depend on the design of new antibacterial agents with high activity and low toxicity. Multi-target methodologies focused on quantitative-structure activity relationships (mt-QSAR), have contributed to rationalize the process of drug discovery, improving the knowledge about the molecular patterns related with antimicrobial activity. Until know, almost all mt-QSAR models have considered the study of biological activity or toxicity separately. Here, we developed a unified mtk-QSBER (multitasking quantitative-structure biological effect relationships) model for simultaneous prediction of anti-enterococci activity and toxicity on laboratory animal and human immune cells. The mtk-QSBER model was created by using artificial neural network (ANN) analysis combined with topological indices, with the aim of classifying compounds as positive (high biological activity and/or low toxicity) or negative (otherwise) under multiple experimental conditions. The mtk-QSBER model correctly classified more than 90% of the whole dataset (more than 10900 cases). We used the model to predict multiple biological effects of the investigational drug BC-3781. Results demonstrate that our mtk-QSBER may represent a new horizon for the discovery of desirable anti-enterococci drugs.

Paper Nr: 13
Title:

Selection of Sensors that Influence Trouble Condition Sign Discovery based on a One-class Support Kernel Machine for Hydroelectric Power Plants

Authors:

Hiroshi Murata, Yasushi Shinohara and Takashi Onoda

Abstract: Trouble conditions rarely occur in the equipment of hydroelectric power plants. Therefore, it is important to find indicator signs for trouble conditions. In a previous study, we proposed a trouble condition sign discovery method, which consists of two detection stages. In the first stage, we can discover trouble condition signs, which are different from the usual condition data. In the second stage, we can monitor aging degradation, with plant experts confirm these trouble condition signs in daily operations. Hence, there is a need to detect these trouble condition signs using a small number of sensors. In this paper, we propose a method for narrowing down the sensors used in trouble condition sign discovery. This paper shows the experimental results of trouble condition sign detection for bearing vibration based on the collected data from different sensors using our proposed method and our previously proposed method. The experimental results show that even if the number of sensors is reduced, our proposed method can find trouble condition signs, which are different from the usual condition data. Therefore, the proposed method may be useful for trouble condition sign discovery in hydroelectric power plants.

Paper Nr: 15
Title:

Automatic Detection of Single Slow Eye Movements and Analysis of their Changes at Sleep Onset

Authors:

Filippo Cona, Fabio Pizza, Federica Provini and Elisa Magosso

Abstract: An algorithm that can automatically identify slow eye movements from the electro-oculogram is presented. The automatic procedure is trained using the visual classification of an expert scorer. The algorithm makes use of both the spectral and morphological signal information to detect single slow eye movements. On the basis of this detection some parameters that characterize the slow eye movements (amplitude, duration, velocity and number) are extracted. A few possible applications of the algorithm are shown by means of a preliminary study: the average patterns of slow eye movements parameters at sleep onset are evaluated for healthy volunteers and for patients affected by obstructive sleep apnea syndrome. Finally, general considerations are drawn regarding the clinical interest of the study.

Paper Nr: 18
Title:

Bayesian versus Neural Network Analysis of Algae Data Population - A New Method to Predict and Analyse Cause and Effect

Authors:

Jen J. Lee, Jorge A. Achcar, Emílio A. C. Barros and Carlos D. Maciel

Abstract: In biology, advanced modelling techniques are needed since there is a mixture of qualitative, linguistics and numerical data on the environmental and biological relationships. Also, experiments and data collecting are expensive and time consuming, so determine which variables are relevant and using inference models less data demanding are highly desirable. In this work, from a set of 200 multivariate data samples of algae population and environmental variables, we propose a Bayesian method to predict compositional population distribution. This is a good application example, since measuring environmental variables are easier to automate, faster and less expensive than population counting that usually involves the need of a large amount of specialized human interaction. An additive log-ratio transformation and a regression model were applied to the data and 255.000 Gibbs samples were simulated using the OPENBUGS software. Also an Artificial Neural Network (ANN) was designed on Matlab to predict the distribution for benchmarking purposes. Both models showed similar prediction performance, but on the Bayesian model an analysis of credible interval of the variables corresponding to the each regression parameters is possible, showing that most of the variables on this study are relevant, which is consistent to the expected results in this case.

Paper Nr: 21
Title:

Evaluating Neuromodulator-controlled Stochastic Plasticity for Learning Recurrent Neural Control Networks

Authors:

Christian W. Rempis, Hazem Toutounji and Frank Pasemann

Abstract: Learning recurrent neural networks as behavior controllers for robots requires measures to guide the learning towards a desired behavior. Organisms in nature solve this problem with feedback signals to assess their behavior and to refine their actions. In line with this, a neural framework is developed where the synaptic learning is controlled by artificial neuromodulators that are produced in response to (undesired) sensory signals. To test this framework and to get a base line to evaluate further approaches, we perform five classical benchmark experiments with a simple random plasticity method. We show that even with this simple plasticity method, behaviors can already be found for all experiments, even for comparably large networks with over 90 plastic synapses. The performance depends strongly on the complexity of the task and less on the chosen network topology. This suggests that controlling learning with neuromodulators is a viable approach that is promising to work also with more sophisticated plasticity methods in the future.

Paper Nr: 22
Title:

The Problem of Organizing and Partitioning Large Data Sets in Learning Algorithms for SOM-RBF Mixed Structures - Application to the Approximation of Environmental Variables

Authors:

José A. Torres, Sergio Martinez, Francisco J. Martinez and Mercedes Peralta

Abstract: The paper presents a technique to partition and sort data in a large training set for building models of envi-ronmental function approximation using RBFs networks. This process allows us to make very accurate ap-proximations of the functions in a time fraction related to the RBF networks classic training proccess. Fur-thermore, this technique avoids problems of buffer overflow in the training algorithm execution. The results obtained proved similar accuracy to those obtained with a classical model in a time substantially less, opening, on the other hand, the way to the parallelization process using GPUs technology.

Paper Nr: 26
Title:

Dynamically Reconfigurable Online Self-organising Fuzzy Neural Network with Variable Number of Inputs for Smart Home Application

Authors:

Anjan Kumar Ray, Gang Leng, T. M. Mcginnity, Sonya Coleman and Liam Maguire

Abstract: A self-organising fuzzy-neural network (SOFNN) adapts its structure based on variations of the input data. Conventionally in such self-organising networks, the number of inputs providing the data is fixed. In this paper, we consider the situation where the number of inputs to a network changes dynamically during its online operation. We extend our existing work on a SOFNN such that the SOFNN can self-organise its structure based not only on its input data, but also according to the changes in the number of its inputs. We apply the approach to a smart home application, where there are certain situations when some of the existing events may be removed or new events emerge, and illustrate that our approach enhances cognitive reasoning in a dynamic smart home environment. In this case, the network identifies the removed and/or added events from the received information over time, and reconfigures its structure dynamically. We present results for different combinations of training and testing phases of the dynamic reconfigurable SOFNN using a set of realistic synthesized data. The results show the potential of the proposed method.

Paper Nr: 27
Title:

Neural Networks Ensemble for Quality Monitoring

Authors:

P. Thomas, M. Noyel, M. C. Suhner, P. Charpentier and A. Thomas

Abstract: Product quality level is a key concept for companies' competitiveness. Different tools may be used to improve quality such as the seven basic quality tools or experimental design. In addition, the need of traceability leads companies to collect and store production data. Our paper aims to show that we can ensure the required quality thanks to an "on line quality approach" based on exploitation of collected data by using neural networks tools. A neural networks ensemble is proposed to classify quality results which can be used in order to prevent defects occurrence. This approach is illustrated on an industrial lacquering process. Results of the neural networks ensemble are compared with the ones obtained with the best neural network classifier.

Paper Nr: 37
Title:

Manifold Learning Approach toward Image Feature-based State Space Construction

Authors:

Yuichi Kobayashi, Ryosuke Matsui and Toru Kaneko

Abstract: This paper presents a bottom-up approach to building internal representation of an autonomous robot under a stand point that the robot create its state space for planning and generating actions only by itself. For this purpose, image-feature-based state space construction method is proposed using LLE (locally linear embedding). The visual feature is extracted from sample images by SIFT (scale invariant feature transform). SOM (Self Organizing Map) is introduced to find appropriate labels of image features throughout images with different configurations of robot. The vector of visual feature points mapped to low dimensional space express relation between the robot and its environment. The proposed method was evaluated by experiment with a humanoid robot collision classification.

Posters
Paper Nr: 25
Title:

Prediction based – High Frequency Trading on Financial Time Series

Authors:

Farhad Kia and Janos Levendovszky

Abstract: In this paper we investigate prediction based trading on financial time series assuming general AR(J) models and mean reverting portfolios. A suitable nonlinear estimator is used for predicting the future values of a financial time series will be provided by a properly trained FeedForward Neural Network (FFNN) which can capture the characteristics of the conditional expected value. In this way, one can implement a simple trading strategy based on the predicted future value of an asset price or a portfolio and comparing it to the current value. The method is tested on FOREX data series and achieved a considerable profit on the mid price. In the presence of the bid-ask spread, the gain is smaller but it still ranges in the interval 2-6 percent in 6 months without using any leverage. FFNNs were also used to predict future values of mean reverting portfolios after identifying them as Ornstein-Uhlenbeck processes. In this way, one can provide fast predictions which can give rise to high frequency trading on intraday data series.

Paper Nr: 33
Title:

Multiplicative Neural Network with Thresholds

Authors:

Leonid Litinskii and Magomed Malsagov

Abstract: The memory of Hopfield-type neural nets is understood as the ground state of the net – a set of configurations providing a global energy minimum. The use of thresholds allows good control over the ground state. It is possible to build multiplicative networks with the degeneracy of the ground state exceeding considerably the dimensionality of the problem (that is, the net memory can be much greater than the dimensionality of the problem). The paper considers the potentials and limitations of the approach.

Paper Nr: 39
Title:

Damaged Letter Recognition Methodology - A Comparison Study

Authors:

Eva Volna, Vaclav Kocian, Michal Janosek, Hashim Habiballa and Vilem Novak

Abstract: The problem of optical character recognition is often solved, not only in the field of artificial intelligence itself, but also in everyday computer usage. We encountered this problem within the industrial project solved for real-life application. Best solver of such a task still remains human brain. Human beings are capable of character recognition even for damaged and highly incomplete images. In this paper, we present alternative softcomputing methods based on application of neural networks and fuzzy logic with evaluated syntax. We proposed a methodology of damaged letters recognition, which was experimentally verified. All experimental results were mutually compared in conclusion. Training and test sets were provided by Company KMC Group, s.r.o.

Paper Nr: 41
Title:

A Spiking Neuron Model based on the Lambert W Function

Authors:

Yevgeniy Bodyanskiy and Artem Dolotov

Abstract: A model of spiking neuron based on the Lambert W function has been proposed. It is shown analytical dependence of spiking neuron firing time on input spikes can be obtained. Though such dependence is rather complex, it still allows of simplifying software implementation of spiking neural networks. It is demonstrated the proposed model software implementation operates faster than one of straightforward propagation of a spike through multiple synapse and soma of spiking neuron.