ICNC 2010 Abstracts


Full Papers
Paper Nr: 25
Title:

A NOVEL ADAPTIVE CONTROL VIA SIMPLE RULE(S) USING CHAOTIC DYNAMICS IN A RECURRENT NEURAL NETWORK MODEL AND ITS HARDWARE IMPLEMENTATION

Authors:

Ryosuke Yoshinaka, Masato Kawashima, Yuta Takamura, Hitoshi Yamaguchi, Naoya Miyahara and Kei-ichiro Nabeta

Abstract: A novel idea of adaptive control via simple rule(s) using chaotic dynamics in a recurrent neural network model is proposed. Since chaos in brain was discovered, an important question, what is the functional role of chaos in brain, has been arising. Standing on a functional viewpoint of chaos, the authors have been proposing that chaos has complex functional potentialities and have been showing computer experiments to solve many kinds of ”ill-posed problems”, such as memory search and so on. The key idea is to harness the onset of complex nonlinear dynamics in dynamical systems. More specifically, attractor dynamics and chaotic dynamics in a recurrent neural network model are introduced via changing a system parameter, ”connectivity”, and adaptive switching between attractor regime and chaotic regime depending surrounding situations is applied to realizing complex functions via simple rule(s). In this report, we will show (1)Global outline of our idea, (2)Several computer experiments to solve 2-dimensional maze by an autonomous robot having a neural network, where the robot can recognize only rough directions of target with uncertainty and the robot has no pre-knowledge about the configuration of obstacles (ill-posed setting), (3)Hardware implementations of the computer experiments using two-wheel or two-legs robots driven by our neuro chaos simulator. Successful results are shown not only in computer experiments but also in practical experiments, (4)Making pseudo-neuron device using semiconductor and opto-electronic technologies, where the device is called ”dynamic self-electro optical effect devices (DSEED)”. They could be ”neuromorphic devices” or even ”brainmorphic devices”.

Paper Nr: 28
Title:

MODELING SKIN BLOOD FLOW - A Neuro-physiological Approach

Authors:

Boris R. M. Kingma, Wim H. Saris and Arjan J. H. Frijns

Abstract: In humans skin blood flow (SBF) plays a major role in body heat loss. Therefore the accuracy of models of human thermoregulation depends for a great deal on their ability to predict skin blood flow. Most SBF-models use body temperatures directly for calculation of skin perfusion. However, humans do not sense temperature directly, yet the information is coded into neuron fire rates. The aim of this study was to investigate whether SBF can be adequately modelled through simulation of temperature sensitive neurons and neuro-physiological pathways of excitation and inhibition. Methods: In this study a mathematical model for SBF was developed based on physiological knowledge on neural thermo-sensitivity and neural pathways. The model was fitted on human experimental data. Mean squared residuals (MSR) were estimated through k-fold cross-validation. Results: The model adequately explains the variance of the measurements (r2=0.91). Furthermore the averaged MSR is close to the natural variation in the measurements (AMSR=0.087 vs. r2=01.080) indicating a small bias. Conclusion: In this study we developed a model for skin perfusion based on physiological evidence on thermo-reception and neural pathways. Given the highly explained variance this study shows that a neuro-physiological approach is applicable for modelling skin blood flow in thermoregulation.

Paper Nr: 34
Title:

AUTOMATIC GAIN CONTROL NETWORKS FOR MULTIDIMENTIONAL VISUAL ADAPTATION

Authors:

S. Furman and Y. Y. Zeevi

Abstract: Processing and analysis of images are implemented in the multidimensional space of visual information representation. This space includes the well investigated dimensions of intensity, color and spatio-temporal frequency. There are, however, additional less investigated dimensions such as curvature, size and depth (for example - from binocular disparity). Along these dimensions, the human visual system (HVS) enhances and emphasizes important image attributes by adaptation and nonlinear filtering. It is interesting and possible to emulate the visual system processing of images along these dimensions, in order to achieve intelligent image processing and computer vision. Sparsely connected, recurrent adaptive sensory neural network (NN), incorporating non-linear interactions in the feedback loops, are presented. Such generic NN exhibit Automatic Gain Control (AGC) model of processing along the visual dimensions. The results are compared with those of psychophysical experiments exhibiting good reproduction of visual illusions.

Paper Nr: 42
Title:

SACCADES GENERATION - From the Visual Input to the Superior Colliculus

Authors:

Wahiba Taouali

Abstract: The superior colliculus is an important structure in the visuomotor pathway of mammals, that is known to be deeply involved in visual saccadic behavior. We present a model of this structure based on biological data, the specificity of which is related to the homogeneity of the underlying substratum of computation. This makes it more suitable to process massive visual flows on a distributed architecture, as it could be requested in a realistic task in autonomous robotics. The model presented here is embedded in the exogenous part of the visual pathway, from the retina to the superior colliculus.

Paper Nr: 44
Title:

NEURAL NETWORK BASED HAMMERSTEIN SYSTEM IDENTIFICATION USING PARTICLE SWARM SUBSPACE ALGORITHM

Authors:

S. Z. Rizvi and H. N. Al-Duwaish

Abstract: This paper presents a new method for modeling of Hammerstein systems. The developed identification method uses state-space model in cascade with radial basis function (RBF) neural network. A recursive algorithm is developed for estimating neural network synaptic weights and parameters of the state-space model. No assumption on the structure of nonlinearity ismade. The proposed algorithm works under the weak assumption of richness of inputs. The problem of modeling is solved as an optimization problem and Particle Swarm Optimization (PSO) is used for neural network training. Performance of the algorithm is evaluated in the presence of noisy data and Monte-Carlo simulations are performed to ensure reliability and repeatability of the identification technique.

Paper Nr: 51
Title:

GRAPHLET DATA MINING OF ENERGETICAL INTERACTION PATTERNS IN PROTEIN 3D STRUCTURES

Authors:

Carsten Henneges and Marc Röttig

Abstract: Interactions between secondary structure elements (SSEs) in the core of proteins are evolutionary conserved and define the overall fold of proteins. They can thus be used to classify protein families. Using a graph representation of SSE interactions and data mining techniques we identify overrepresented graphlets that can be used for protein classification. We find, in total, 627 significant graphlets within the ICGEB Protein Benchmark database (SCOP40mini) and the Super-Secondary Structure database (SSSDB). Based on graphlets, decision trees are able to predict the four SCOP levels and SSSDB (sub)motif classes with a mean Area Under Curve (AUC) better than 0.89 (5-fold CV). Regularized decision trees reveal that for each classification task about 20 graphlets suffice for reliable predictions. Graphlets composed of five secondary structure interactions are most informative. Finally, we find that graphlets can be predicted from secondary structure using decision trees (5-fold CV) with a Matthews Correlation Coefficient (MCC) reaching up to 0.7.

Paper Nr: 59
Title:

DISCOVERING CORTICAL ALGORITHMS

Authors:

Atif G. Hashmi and Mikko H. Lipasti

Abstract: We describe a cortical architecture inspired by the structural and functional properties of the cortical columns distributed and hierarchically organized throughout the mammalian neocortex. This results in a model which is both computationally efficient and biologically plausible. The strength and robustness of our cortical architecture is ascribed to its distributed and uniformly structured processing units and their local update rules. Since our architecture avoids complexities involved in modeling individual neurons and their synaptic connections, we can study other interesting neocortical properties like independent feature detection, feedback, plasticity, invariant representation, etc. with ease. Using feedback, plasticity, object permanence, and temporal associations, our architecture creates invariant representations for various similar patterns occurring within its receptive field. We trained and tested our cortical architecture using a subset of handwritten digit images obtained from the MNIST database. Our initial results show that our architecture uses unsupervised feedforward processing as well as supervised feedback processing to differentiate handwritten digits from one another and at the same time pools variations of the same digit together to generate invariant representations.

Paper Nr: 61
Title:

CORTICAL RHYTHMS INDUCED BY TMS STIMULATION - Analysis with a Neural Mass Model

Authors:

Filippo Cona and Melissa Zavaglia

Abstract: Knowledge of cortical rhythms represents an important aspect of modern neuroscience, to understand how the brain realizes its functions. Recent data suggest that different regions in the brain may exhibit distinct rhythms when perturbed by Transcranial Magnetic Stimulation (TMS) (Rosanova et al., 2009) and that these rhythms can change due to the connectivity among regions. In this context, neural mass models can be very useful to simulate specific aspects of electrical brain activity and, above all, to analyze and identify the overall frequency content of EEG in a cortical region of interest (ROI). In this work we implemented a model of connectivity among cortical regions (Ursino, Cona and Zavaglia, 2010) to fit the impulse responses in three ROIs during an experiment of TMS stimulation. In particular we investigated Brodmann Area (BA) 19 (occipital lobe), BA 7 (parietal lobe) and BA 6 (frontal lobe). Results show that the model can reproduce the natural rhythms of the three regions quite well, acting on a few internal parameters. Moreover, model can explain most rhythm changes induced by stimulation of another region, by using just a few long-range connectivity parameters among ROIs.

Paper Nr: 65
Title:

USING SLOW FEATURE ANALYSIS TO IMPROVE THE REACTIVITY OF A HUMANOID ROBOT'S SENSORIMOTOR GAIT PATTERN

Authors:

Sebastian Höfer and Manfred Hild

Abstract: This paper presents an approach for increasing the reactivity of a humanoid robot’s gait, incorporating Slow Feature Analysis (SFA), an unsupervised learning algorithm issuing from the domain of theoretical biology. The main objective of this work is to find a means to detect disturbances in the gait pattern at an early stage without losing stability. Another goal is to investigate the general potential of SFA for using it within sensorimotor loops which to our knowledge has not been considered until now. The application of SFA within sensorimotor loops is motivated by pointing out its relation to second-order Volterra filters. Our experiments show that the overall reactivity of the gait pattern increases without any profound loss in stability, and that SFA appears to be suitable for the usage even at such levels of sensorimotor control that are directly involved into motor activity regulation.

Paper Nr: 72
Title:

A MINIMAL CONTROL SCHEMA FOR GOAL-DIRECTED ARM MOVEMENTS BASED ON PHYSIOLOGICAL INTER-JOINT COUPLINGS

Authors:

Till Bockemühl and Volker Dürr

Abstract: Substantial evidence suggests that nervous systems simplify motor control of complex body geometries by use of higher level functional units, so called motor primitives or synergies. Although simpler, such high level functional units still require an adequate controller. In a previous study, we found that kinematic inter-joint couplings allow the extraction of simple movement synergies during unconstrained 3D catching movements of the human arm and shoulder girdle. Here, we show that there is a bijective mapping between movement synergy space and 3D Cartesian hand coordinates within the arm’s physiological working range. Based on this mapping, we propose a minimal control schema for a 10-DoF arm and shoulder girdle. All key elements of this schema are implemented as artificial neural networks (ANNs). For the central controller, we evaluate two different ANN architectures: a feed-forward network and a recurrent Elman network. We show that this control schema is capable of controlling goal-directed movements of a 10-DoF arm with as few as five hidden units. Both controller variants are sufficient for the task. However, end-point stability is better in the feed-forward controller.

Paper Nr: 75
Title:

SMART GROWING CELLS

Authors:

Hendrik Annuth and Christian-A. Bohn

Abstract: General unsupervised learning or self-organization places n-dimensional reference vectors in order to match the distribution of samples in an n-dimensional vector space. Beside this abstract view on self-organization there are many applications where training — focused on the sample distribution only — does not lead to a satisfactory match between reference cells and samples. Kohonen’s self-organizing map, for example, overcomes pure unsupervised learning by augmenting an additional 2D topology. And although pure unsupervised learning is restricted therewith, the result is valuable in applications where an additional 2D structure hidden in the sample distribution should be recognized. In this work, we generalize this idea of application-focused trimming of ideal, unsupervised learning and reinforce it through the application of surface reconstruction from 3D point samples. Our approach is based on Fritzke’s growing cells structures (GCS) (Fritzke, 1993) which we extend to the smart growing cells (SGC) by grafting cells by a higher-level intelligence beyond the classical distribution matching capabilities. Surface reconstruction with smart growing cells outperforms most neural network based approaches and it achieves several advantages compared to classical reconstruction methods.

Short Papers
Paper Nr: 14
Title:

SVM-BASED HUMAN DETECTION COMBINING SELF-QUOTIENT ε-FILTER AND HISTOGRAMS OF ORIENTED GRADIENTS

Authors:

Mitsuharu Matsumoto

Abstract: This paper describes a noise robust SVM-based human detection combining self-quotient ε-filter (SQEF) and histograms of oriented gradients (HOG). Although human detection combining HOG and SVM is a powerful approach, as it uses local intensity gradients, it is difficult to handle noise corrupted images. To handle noise corrupted images, we introduce self-quotient ε-filter (SQEF), and implement it in human detection combining HOG and SVM. SQEF is an advanced self-quotient filter (SQF), and can clearly extract features from the images not only when they have illumination variations but also when they are corrupted with noise. The new approach gives a robust human detection from noise corrupted images using the data trained by intact images without noise.

Paper Nr: 20
Title:

CLASSIFICATION AND CLUSTERING OF BRAIN INJURIES FROM MOTION DATA OF PATIENTS IN A VIRTUAL REALITY ENVIRONMENT

Authors:

Uri Feintuch

Abstract: Virtual Reality (VR) has been found to be an effective rehabilitation tool for brain injury patients. We show that motion data from these VR sessions can be effectively used to both cluster and classify patients according to types of injury. Neural Network and other tools were used to differentially classify patients with traumatic brain injury, cerebral vascular accident (stroke) with and without spatial neglect and healthy individuals solely from the motion data. Clustering techniques also successfully duplicated the classification division. These results have potential implications for scientific research, automated diagnosis and integrated individually adaptive therapies in the virtual reality technology.

Paper Nr: 23
Title:

ESTIMATION OF QUANTUM TIME LENGTH FOR ROUND-ROBIN SCHEDULING ALGORITHM USING NEURAL NETWORKS

Authors:

Omar AlHeyasat and Randa Herzallah

Abstract: The quantum time length is usually taken as a fixed value in all applications that use Round Robin (RR) scheduling algorithm. The determination of the optimal length of the quantum that results in a small average turn around time is very complicated because of the unknown nature of the tasks in the ready queue. The round robin algorithm becomes very similar to the first in first served algorithm if the quantum length is large. On the other hand, high context switch results for small values of quantum length which might cause central processing unit (CPU) thrashing. In this paper we propose a new RR scheduling algorithm based on using neural network models for predicting the optimal quantum length that yields minimum average turn around time. The quantum length is taken to be a function of the service time of the various jobs available in the ready queue. This in contrast to the traditional methods of using fixed quantum length is shown to give better results and to minimize the average turnaround time for almost any collection of jobs in the ready queue.

Paper Nr: 27
Title:

THE LIQUID STATE MACHINE IS NOT ROBUST TO PROBLEMS IN ITS COMPONENTS BUT TOPOLOGICAL CONSTRAINTS CAN RESTORE ROBUSTNESS

Authors:

Hananel Hazan and Larry Manevitz

Abstract: The Liquid State Machine (LSM) is a method of computing with temporal neurons, which can be used amongst other things for classifying intrinsically temporal data directly unlike standard artificial neural networks. It has also been put forward as a natural model of certain kinds of brain functions. There are two results in this paper: (1) We show that the LSM as normally defined cannot serve as a natural model for brain function. This is because they are very vulnerable to failures in parts of the model. This result is in contrast to work by Maass et al which showed that these models are robust to noise in the input data. (2) We show that specifying certain kinds of topological constraints (such as "small world assumption"), which have been claimed are reasonably plausible biologically, can restore robustness in this sense to LSMs.

Paper Nr: 29
Title:

RECURRENT NEURAL NETWORK WITH SOFT 'WINNER TAKES ALL' PRINCIPLE FOR THE TSP

Authors:

Paulo Henrique Siqueira

Abstract: This paper shows the application of Wang’s Recurrent Neural Network with the 'Winner Takes All' (WTA) principle in a soft version to solve the Traveling Salesman Problem. In soft WTA principle the winner neuron is updated at each iteration with part of the value of each competing neuron and some comparisons with the hard WTA are made in this work with instances of the TSPLIB (Traveling Salesman Problem Library). The results show that the soft WTA guarantees equal or better results than the hard WTA in most of the problems tested.

Paper Nr: 30
Title:

INTERACTIONS BETWEEN HEMISPHERES WHEN DISAMBIGUATING AMBIGUOUS HOMOGRAPH WORDS DURING SILENT READING

Authors:

Zohar Eviatar, Hananel Hazan and Larry Manevitz

Abstract: A model of certain aspects of the cortex related to reading is developed corresponding to ongoing exploration of psychophysical and computational experiments on how the two hemispheres work in humans. The connectivity arrangements between modelled areas of orthography, phonology and semantics are according to the theories of Eviatar and Peleg, in particular with distinctions between the connectivity in the right and left hemisphere. The two hemispheres are connected and interact both in training and testing in a reasonably "natural" way. We found that the RH (right hemisphere) serves to maintain alternative meanings under this arrangement longer than the LH for homophones. This corresponds to the usual theories (about homographs) while, surprisingly, the LH maintains alternative meanings longer then the RH for heterophones. This allows the two hemispheres, working together to resolve ambiguities regardless of when the disambiguating information arrives. Human experiments carried out subsequent to these results bear this surprising result out.

Paper Nr: 33
Title:

IMPLICIT SEQUENCE LEARNING - A Case Study with a 4–2–4 Encoder Simple Recurrent Network

Authors:

Stefan Glüge

Abstract: Without any doubt the temporal order inherent in a task is an important issue during human learning. Recurrent neural networks are known to be a useful tool to model implicit sequence learning. In terms of the psychology of learning, recurrent networks might be suitable to build a model to reproduce the data obtained from experiments with human subjects. Such model should not just reproduce the data but also explain it and further make verifiable predictions. Therefore, one basic requirement is an understanding of the processes in the network during learning. In this paper, we investigate how (implicitly learned) temporal information is stored/represented in a simple recurrent network. To be able to study detailed effects we use a small network and a standard encoding task for this study.

Paper Nr: 35
Title:

PROBABILISTIC NEURAL NETWORKS FOR CREDIT RATING MODELLING

Authors:

Petr Hájek

Abstract: This paper presents the modelling possibilities of probabilistic neural networks to a complex real-world problem, i.e. credit rating modelling. First, current approaches in credit rating modelling are introduced. Then, probabilistic neural networks are designed to classify US companies and municipalities into rating classes. The input variables are extracted from financial statements and statistical reports in line with previous studies. These variables represent the inputs of probabilistic neural networks, while the rating classes from Standard&Poor’s and Moody’s rating agencies stand for the outputs. Classification accuracies, misclassification costs, and the contributions of input variables are studied for probabilistic neural networks compared to other neural networks models. The results show that the rating classes assigned to bond issuers can be classified accurately with probabilistic neural networks using a limited subset of input variables.

Paper Nr: 36
Title:

HOW CAN NEURAL NETWORKS SPEED UP ECOLOGICAL REGIONALIZATION FRIENDLY? - Replacement of Field Studies by Satellite Data using RBFs

Authors:

Manolo Cruz, Moisés Espínola and Rosa Ayala

Abstract: The aim of this work is to present an application of the Radial Basis Functions Nets (RBFs) for simplifying and reducing the cost of ecological regionalization. The process speeds up and replaces the classic means of obtaining ecological variables through field studies. The radial basis function networks were applied to estimate field data remotely, using data captured by the Landsat satellite and correlating it with ecological variables in order to substitute for them in the regionalization process. This approach substantially reduces the time and cost of ecological regionalization, limiting field studies and automating the generation of the ecological variables. The technique could be applied without restriction to map vegetation in any other area for which satellite coverage exists.

Paper Nr: 48
Title:

FEATURE CLUSTERING WITH SELF-ORGANIZING MAPS AND AN APPLICATION TO FINANCIAL TIME-SERIES FOR PORTFOLIO SELECTION

Authors:

Bruno Silva and Nuno Marques

Abstract: The portfolio selection is an important technique for decreasing the risk in the stock investment. In the portfolio selection, the investor’s property is distributed for a set of stocks in order to minimize the financial risk in market downturns. With this in mind, and aiming to develop a tool to assist the investor in finding balanced portoflios, we achieved a generic method for feature clustering with Self-OrganizingMaps (SOM). The ability of neural networks to discover nonlinear relationships in input data makes them ideal for modeling dynamic systems as the stock market. The method proposed makes use the remarkable visualization capabilities of the SOM, namely the Component Planes, to detect non-linear correlations between features. An appropriate metric - the improved Rv coefficient - is also proposed to compare Component Planes and generate a distance matrix between features, after which an hierarchical clustering method is used to obtain the clusters of features. Results obtained are empirically sound, although at this moment we do not provide mathematical comparisons with other methods. Results also show that feature clustering with the SOM presents itself as a viable method to cluster time-series.

Paper Nr: 53
Title:

SYSTEM IDENTIFICATION BASED ON MULTI-KERNEL LEAST SQUARES SUPPORT VECTOR MACHINES (MULTI-KERNEL LS-SVM)

Authors:

Mounira Tarhouni and Kaouther Laabidi

Abstract: This paper develops a new approach to identify nonlinear systems. A Multi-Kernel Least Squares Support Vector Machine (Multi-Kernel LS-SVM) is proposed. The basic LS-SVM idea is to map linear inseparable input data into a high dimensional linear separable feature space via a nonlinear mapping technique (kernel function) and to carry out linear classification or regression in feature space. The choice of kernel function is an important task which is related to the system nonlinearity degrees. The suggested approach combines several kernels in order to take advantage of their performances. Two examples are given to illustrate the effectiveness of the proposed method.

Paper Nr: 58
Title:

INVERSE PROBLEMS IN LEARNING FROM DATA

Authors:

Věra Kůrková

Abstract: It is shown that application of methods from theory of inverse problems to learning from data leads to simple proofs of characterization of minima of empirical and expected error functionals and their regularized versions. The reformulation of learning in terms of inverse problems also enables comparison of regularized and non regularized case showing that regularization achieves stability by merely modifying output weights of global minima. Methods of theory of inverse problems lead to choice of reproducing kernel Hilbert spaces as suitable ambient function spaces.

Paper Nr: 62
Title:

NEURAL NETWORK BASED CONTROLLER FOR NONLINEAR AUTOMATIC GENERATION CONTROL

Authors:

S. Z. Rizvi

Abstract: This paper presents an Artificial Neural Network (ANN) based controller design for nonlinear multivariable systems. The proposed method uses a novel algorithm for using and training a radial basis function (RBF) neural network based controller. The training algorithm makes sure that it does not violate any constraints on the inputs or outputs. Trajectory tracking results are presented for the challenging problem of nonlinear single area Automatic Generation Control (AGC) power system. Both linear and nonlinear cases are considered and robustness of the controller is tested as well.

Paper Nr: 66
Title:

WANN-TAGGER - A Weightless Artificial Neural Network Tagger for the Portuguese Language

Authors:

Hugo C. C. Carneiro

Abstract: Weightless Artificial Neural Networks have proved to be a promising paradigm for classification tasks. This work introduces the WANN-Tagger, which makes use of weightless artificial neural networks for labelling Portuguese sentences, tagging each of its terms with its respective part-of-speech. A first experimental evaluation using the CETENFolha corpus indicates the usefulness of this paradigm and shows that it outperforms traditional feedforward neural networks in both accuracy and training time, and also that it is competitive in accuracy with the Hidden Markov Model in some cases. Additionally, WANN-Tagger shows itself capable of incrementally learning new tagged sentences during runtime.

Paper Nr: 71
Title:

LEARNING IN BIOLOGICAL NEUROPROCESSORS USING A CENTER OF AREA METHOD

Authors:

José M. Ferrández, Victor Lorente, Félix de la Paz and José Manuel Cuadra

Abstract: Learning in a biological neuroprocessor is analyzed using human neuroblastoma cultures and a center of area method in order to guide a robot to follow the light or the brightest area in a limited scenario. The main setup consists in an inverted microscope where a multielectrode array is attached with the biological cultures. This elements amplifies and send the weak neural signals to a D/A card where analyzing process is achieved, computing the movement of the robot, that is remotely linked to this computer. The robot also sends the a picture of the scenario to the computer in order to stimulate the culture with a center of area scheme. In this paper, it is shown that learning is possible in this culture, and guiding the robot to a desired goal is a feasible task.

Paper Nr: 73
Title:

AN EFFICIENT IMPLEMENTATION OF A REALISTIC SPIKING NEURON MODEL ON AN FPGA

Authors:

Dominic Just and Jeferson F. Chaves

Abstract: Hardware implementations of spiking neuron models have been studied over the years mainly in researches focused on bio-inspired systems and computational neuroscience. This introduced considerable challenges for researchers particularly in terms of the requirements to realise a efficient embedded solution which may provide artificial devices adaptability and performance in real-time environment. Thus, programmable hardware was widely used as a model for the adaptable requirements of neural networks. From this perspective, this paper describes an efficient implementation of a realistic spiking neuron model on a Field Programmable Gate Array (FPGA). A network consisting of 10 Izhikevich’s neurons was produced, in a low-cost and low-density FPGA. It operates 100 times faster than in real time, and the perspectives of these results in newer models of FPGAs are promising.

Paper Nr: 77
Title:

TRANSPOSING SIMULATED SELF-ORGANIZING ROBOTS INTO REALITY USING THE PLUG&LEARN ARCHITECTURE

Authors:

Frank Güttler, Wolfgang Rabe and Jörn Hoffmann

Abstract: Simulations for robots like the robot simulator LPZROBOTS allow a fast proof of theoretical concepts using self-organizing neural networks. This publication presents a hardware platform as a solution to transpose these theoretical results to real robots without the time consuming reimplementation of algorithms and without the loss of computational power a standard desktop PC offers. This is shown by the example of the THREECHAINED TWOWHEELED robot which gains embodiment and shows the same emergent behaviour in comparison to the simulated counterpart.

Paper Nr: 78
Title:

AN ACTION-TUNED NEURAL NETWORK ARCHITECTURE FOR HAND POSE ESTIMATION

Authors:

Giovanni Tessitore

Abstract: There is a growing interest in developing computational models of grasping action recognition. This interest is increasingly motivated by a wide range of applications in robotics, neuroscience, HCI, motion capture and other research areas. In many cases, a vision-based approach to grasping action recognition appears to be more promising. For example, in HCI and robotic applications, such an approach often allows for simpler and more natural interaction. However, a vision-based approach to grasping action recognition is a challenging problem due to the large number of hand self-occlusions which make the mapping from hand visual appearance to the hand pose an inverse ill-posed problem. The approach proposed here builds on the work of Santello and co-workers which demonstrate a reduction in hand variability within a given class of grasping actions. The proposed neural network architecture introduces specialized modules for each class of grasping actions and viewpoints, allowing for a more robust hand pose estimation. A quantitative analysis of the proposed architecture obtained by working on a synthetic data set is presented and discussed as a basis for further work.

Paper Nr: 80
Title:

AN EXTREME LEARNING MACHINE CLASSIFIER FOR PREDICTION OF RELATIVE SOLVENT ACCESSIBILITY IN PROTEINS

Authors:

Saras Saraswathi

Abstract: A neural network based method called Sparse-Extreme Learning Machine (S-ELM) is used for prediction of Relative Solvent Accessibility (RSA) in proteins. We have shown that multiple-fold gains in speed of processing by S-ELM compared to using SVM for classification, while accuracy efficiencies are comparable to literature. The study indicates that using S-ELM would give a distinct advantage in terms of processing speed and performance for RSA prediction.

Paper Nr: 81
Title:

PROTEIN SECONDARY STRUCTURE PREDICTION USING KNOWLEDGE-BASED POTENTIALS

Authors:

Saras Saraswathi and Robert L. Jernigan

Abstract: A novel method is proposed for predicting protein secondary structure using data derived from knowledge based potentials and Neural Networks. Potential energies for amino acid sequences in proteins are calculated using protein structures. An Extreme Learning Machine classifier (ELM-PSO) is used to model and predict protein secondary structures. Classifier performance is maximized using the Particle Swarm Optimization algorithm. Preliminary results show improved results.

Paper Nr: 85
Title:

DECODING SSVEP RESPONSES USING TIME DOMAIN CLASSIFICATION

Authors:

Nikolay V. Manyakov, Nikolay Chumerin and Adrien Combaz

Abstract: In this paper, we propose a new time domain method for decoding the steady-state visual evoked potential recorded while the subject is looking at stimuli flickering with constant frequencies. Using several such stimuli, with different frequencies, a brain-computer interface can be built. We have assessed the influence of the number of electrodes on the decoding accuracy. A comparison between active wet- and bristle dry electrodes were made. The dependence between accuracy and the length of the EEG interval used for decoding was shown.

Paper Nr: 92
Title:

DOES IT EXIST A LINK BETWEEN PERFORMANCE AND PARIETAL CORTEX ACTIVITY IN SURGICAL TASKS?

Authors:

G. Paggetti, Y.-C. Lin and G. Menegaz

Abstract: This pilot study would to explore the ideas of a possible correlation between the goodness of surgical performance in robotic assisted minimally invasive surgery (MIS) and posterior parietal cortex (PPC) activity. This cortical area is known to be involved in stereoscopic vision (Sakata et al., 1997), visual control of eye movements and hand-eye co-ordination (Shikata et al., 1996). This issue is of great interest because robotic assisted surgery provides the surgeon with a stereoscopic view of the operative field combined with aligned motor-visual axes and mechatronically controlled instruments. In this contribution, we conduct an exploratory experiment aiming at investigating the hypothesis of a correlation between the performance in reached in a surgically relevant task and the activation of PPC channels as revealed by the fNIRS measurements. First results are very promising and suggest the occurrence of a link between performance and channel activation.

Posters
Paper Nr: 13
Title:

PARAMETER SETTING OF NOISE REDUCTION FILTER USING SPEECH RECOGNITION SYSTEM

Authors:

Tomomi Abe

Abstract: This paper describes parameter setting of noise reduction filter using speech recognition system. Parameter setting problem is usually solved by maximization or minimization of some objective evaluation functions such as correlation and statistical independence. However, when we consider a single-channel noisy signal, it is difficult to employ such objective functions. It is also difficult to employ them when we consider impulsive noise because its duration is very small to use this assumption. To solve the problems, we directly use a speech recognition system as evaluation function for parameter setting. As an example, we employ time-frequency e-filter and Julius as a filtering system and a speech recognition system, respectively. The experimental results show that the proposed approach has a potential to set the parameter in unknown environments.

Paper Nr: 22
Title:

PROLONGATION RECOGNITION IN DISORDERED SPEECH USING CWT AND KOHONEN NETWORK

Authors:

Ireneusz Codello and Wiesława Kuniszyk-Jóźkowiak

Abstract: Automatic disorder recognition in speech can be very helpful for the therapist while monitoring therapy progress of the patients with disordered speech. In this article we focus on prolongations. We analyze the signal using Continuous Wavelet Transform with 22 bark scales, we divide the result into vectors (using windowing) and then we pass such vectors into Kohonen network. We have increased the recognition ratio from 54% to 81% by adding a modification into the network learning process as well as into CWT computation algorithm. All the analysis was performed and the results were obtained using the authors’ program – “WaveBlaster”. It is very important that the recognition ratio above 80% was obtained by a fully automatic algorithm (without a teacher). The presented problem is part of our research aimed at creating an automatic prolongation recognition system.

Paper Nr: 43
Title:

A PSO-TRAINED ADAPTIVE NEURO-FUZZY INFERENCE SYSTEM FOR FAULT CLASSIFICATION

Authors:

Haris M. Khalid, S. Z. Rizvi and Lahouari Cheded

Abstract: When a fault occurs during an industrial inspection, workmen have to manually find the location and type of the fault in order to remove it. It is often difficult to accurately find the location and type of fault. Hence, development of an offline intelligent fault diagnosis system for process control industry is of great importance since successful detection of fault is a precursor to fault isolation using corrective actions. This paper presents a novel hybrid Particle Swarm Optimization (PSO) and Subtractive Clustering (SC) based Neuro-Fuzzy Inference System (ANFIS) designed for fault detection. The proposed model uses the PSO algorithm to find optimal parameters for (SC) based ANFIS training. The developed PSO-SC-ANFIS scheme provides critical information about the presence or absence of a fault. The proposed scheme is evaluated on a laboratory scale benchmark two-tank process. Leakage fault is detected and results are presented at the end of the paper showing successful diagnosis of most incipient faults when subjected to a fresh set of data.

Paper Nr: 47
Title:

NEURAL NETWORKS IN COMBUSTION SIMULATIONS

Authors:

Lars Frank Große and Franz Joos

Abstract: The design process of commercially available combustion engines is often based on real experiments which is expensive concerning to fuel consumption, men power and environmental pollution. It is possible to replace complex experiments by computer simulations. The prediction of the velocity field, the mixing process of fuel and oxidiser and the temperature field is a wide range of research subjects. In case of turbulent flow simulations with combustion the chemical reactions and the coupling have to be calculated at the same time. With regard to computer time the used chemical reaction mechanism has a big influence on the performance of the whole simulation. Therefore optimisation procedures often improve the representation of the chemistry. The suggestion made in this paper, is the use of artificial neuronal networks for approximation of complex chemistry in turbulent combustion simulations.

Paper Nr: 49
Title:

DEVELOPING MULTIVARIATE MODELS TO PREDICT ABNORMAL STOCK RETURNS - Using Cross-sectional Differences to Identify Stocks with Above Average Return Expectations

Authors:

Alwyn J. Hoffman

Abstract: This paper describes the development of multivariate models used to identify stocks with above average return expectations. While most other research involving the development of stock return models involves time-series prediction of future returns, this paper focuses on the modelling of cross-sectional differences between stocks. The primary measure used in this paper to evaluate potential predictors of future stock returns is based on sorted category returns, an approach that was previously applied to NYSE listed stocks; in this paper the same approach is applied to stocks listed on the JSE. This measure is used to identify a number of fundamental and technical indicators that differentiates between high and low performing stock categories. Linear and non-linear multivariate models are subsequently developed, utilising these indicators to improve prediction performance. It is demonstrated that much of the useful stock return behaviour is present in the extremes of the population, that significant differences exist between different size categories, and that different aspects of stock behaviour is exposed using appropriate measures for portfolio returns. Portfolio performance results achieved using individual indicators as well as multivariate models are reported and compared with previously published results, and planned future work to improve on the results is discussed.

Paper Nr: 57
Title:

PERFORMANCE COMPARISON OF A BIOLOGICALLY INSPIRED EDGE DETECTION ALGORITHM ON CPU, GPU AND FPGA

Authors:

Patrick Dempster and Thomas M. Mcginnity

Abstract: Implementation of Spiking neural networks (SNNs) are becoming an important computational platform for bio-inspired engineers and researchers. However, as networks increase in size towards the biological scale. Ever increasing simulation times are becoming a substantial problem. Efforts to simulate this problem have been many and varied. Modern Graphic Processing Units (GPUs) are increasingly being employed as a platform, whose parallel array of streaming multiprocessors (SMs) allow many thousands of lightweight threads to run. This paper presents a GPU implementation of an SNN application which performs edge detection. The approach is then compared with an equivalent implementations on an Intel Xeon CPU and an FPGA system. The GPU approach was found to provide a speed up of 1.37 times over the FPGA version and an increase of 23.49 times when compared with the CPU based software simulation.

Paper Nr: 69
Title:

SELF ORGANIZING NEURAL NETWORK APPLICATION FOR SKIN COLOR SEGMENTATION

Authors:

David González-Ortega, F. J. Díaz-Pernas, M. Antón-Rodríguez, M. Martínez-Zarzuela and I. de la Torre-Díez

Abstract: In this paper, we present a Fuzzy ART (Adaptive Resonance Theory) neural network application for skin color segmentation using the chromaticity components of the TSL color space. The Fuzzy ART networks deal with the stability-plasticity dilemma and they can be applied to color image segmentation, particularly to skin color segmentation. The developed application has three modes: parameter setting, skin color filter creation, and skin color filter performance. Many parameters can be tuned to create proper skin color filters from manually selected skin regions in an image. A skin color filter is a LUT (Look-Up Table) that gives each color in the RGB color space, one of two different outputs, skin or non-skin color. The performance of different skin color filters can be compared with the application. A skin color filter can be used to make robust real-time skin color segmentation in video sequences captured by a webcam.

Paper Nr: 70
Title:

MULTI MOTHER WAVELET NEURAL NETWORK BASED ON GENETIC ALGORITHM FOR 1D AND 2D FUNCTIONS’ APPROXIMATION

Authors:

Mejda Chihaoui

Abstract: This paper presents a new wavelet-network-based technique for 1D and 2D functions’ approximation. Classical training algorithms start with a predetermined network structure which can be either insufficient or overcomplicated. Furthermore, the resolutions of wavelet networks training problems by gradient are characterized by their noticed inability to escape of local optima. The main feature of this technique is that it avoids both insufficiency and local minima by including genetic algorithms. Simulation results are demonstrated to validate the generalization ability and efficiency of the proposed Multi Mother Wavelet Neural Network based on genetic algorithms.

Paper Nr: 76
Title:

A HYBRID EXPERT SYSTEM BASED ON NEURAL NETWORKS AND FUZZY LOGIC FOR FAULT IDENTIFICATION IN ELECTRIC POWER SUBSTATIONS

Authors:

Daniel da Silva Gazzana, Mario Orlando Oliveira, Arturo Suman Bretas and Andre Lerm

Abstract: This paper presents a novel approach for on-line fault identification in an Electric Power Substation (EPS). The proposed methodology is based on signal processing techniques allied with a Fuzzy Logic and Artificial Neural Network. The test electric system was rigorously built in an electromagnetic transient numerical simulator, named Alternative Transient Program (ATP), conformably to the needs presented by a Thermoelectric Generation Plant of 711 MW - 230 kV, located in southern Brazil. Simulated test cases demonstrate the generalization capability of the developed hybrid Expert System based on Neural Networks and Fuzzy Logic, now utilized in a Southern Brazilian Utility.