
Interuniversity
Attraction
Poles "Dynamical
Systems and Control: 
Study Day
Study
Day of the Interuniversity Attraction Poles (IAP) V/22
Tuesday 26 November 2002
Hôtel Mercure  Avenue de Lauzelle 61, 1348 LouvainlaNeuve

PROGRAMME
09h30 Plenary
Lecture I
“Similarity
in graphs. Application to web searching and to automatic synonym extraction”
by Professor Vincent Blondel
10h30 A Tribute to
Frank Callier
10h45 Poster Session
I  Newcomers within the 7 teams of the IAP Network
12h00 Meeting of
the IAP V/22 promoters and representatives of O.S.T.C.
12h30 Lunch
14h00 Poster Session
II – Traditional contributed poster session
15h30 Coffee break
16h00 Plenary Lecture
II
“Coordinated
control of multiagent systems”
by Professor Luc Moreau
17h00
Closing.
Plenary Lecture I
Similarity
in graphs: Application to web searching and to automatic synonym extraction
Professor Vincent Blondel
UCLINMA
Abstract
We introduce a new concept of similarity between nodes in graphs and describe applications of this concept to the automatic extraction of synonyms in a dictionary and to web searching.
In the graph of the web, every webpage is a node and there is a directed edge from node A to node B if page A points to page B. Some of today's most efficient search engines exploit this graph structure by looking for the most “important” nodes in a subgraph constructed from the given query. Different methods exist for identifying important nodes in a graph. One such recent method proposed by Kleinberg assigns a hub score and an authority score to every node. Pages with a high hub score are expected to be good “navigation pages”, and those with a high authority score good “content pages”. These scores are mutually reinforcing and are obtained as the result of a converging iterative process.
In our talk, we introduce a new concept of similarity between nodes in graphs. Associated to two directed graphs GA and GB we construct a similarity matrix S whose entry (i,j) expresses how similar vertex i (in GA) is to vertex j (in GB). The potential applications of our similarity matrix for information retrieval purposes are manifold. For a particular graph GA with two nodes the similarity matrix gives the hub and authority webpage scores of Kleinberg; and for a three nodes graph the similarity matrix is appropriate for searching synonyms in a dictionary graph. For this last applications, we report results obtained on the English dictionary “Webster” and on the French dictionary “Le Petit Robert”.
The concept of similarity matrix is joint work with Paul Van Dooren. The application of this concept to synonym extraction is joint work with Pierre Senellart.
***
Plenary Lecture II
Coordinated
control of multiagent systems
Professor Luc Moreau
RUGSYSTeMS
Abstract
This talk focuses upon the control of groups of agents (mobile robots, underwater vehicles, etc.). Several formation control maneuvers are considered, where the agents are required to assume a prescribed position with respect to each other. The design methodology presented in this talk is of a decentralized nature.
Instead of having a leader in the group, all agents are considered to be equal, thus increasing the robustness of the group with respect to failure of individuals. Special attention goes to the issue of bandwidth constraints, which limits the communication between individual agents.
***
Poster Session I
Augmented barriers for selfscaled cones
Michel Baes
UCLINMA
Abstract
We handle a recently studied object in interiorpoint method for convex programming: the augmented barrier, which is a selfconcordant barrier $F$ for a regular cone $\K$ augmented by a quadratic form $\langle Qx,x\rangle$. Strangely enough, the problem of finding the analytic centre of such a function appears to be almost universal. What is more, when we deal with the positive orthant or with the psd cone, if $Q\K\subseteq \K$, we can solve that problem with a complexity depending only on the parameter of $F$.
We think it will be possible to generalise these results to all the class of socalled selfscaled cones. We have already achieved a complete treatment of the Lorentz cone case, getting, as side results, an explicit caracterisation of the automorphisms of the Lorentz cone and a complete complexity study of the problem of minimisation of a quadratic function on the unit ball.
*****
Dynamics and computation; Complexity and control
JeanCharles Delvenne
UCLINMA
Abstract
Two main lines are currently followed.
First, we would like to define under which conditions a family of dynamical systems can be chosen as a model of computation, and on the other hand find which properties of dynamical systems are undecidable.
Second, we attempt to answer the question: which systems are difficult to control? More precisely, we wish to define a measure of complexity of control, with the help of computer science.
*****
Model reduction of mechanical systems
Damien Lemmonier
UCLINMA
Abstract
The current objective of my research is to provide methods of constructing reduced models of large scale linear mechanical systems that preserve the “secondorder structure” of the original system. Because the systems we deal with are LTI, we could use the popular “balanced truncation” reduction method, which gives nearly optimal errors (in some sense). But this method does generally not preserve the mechanical structure. I therefore intend to develop methods that approximate the balanced truncation in such a way that the reduced model is a mechanical one and that the induced reduction error remains small.
Keywords: Model reduction, Balanced truncation, Hankel singular values, Projectors, Operators theory.
*****
The influence of cognitive cues on the predictive control of eye movements
JeanJacques Orban
UCLINMA
Abstract
The orientation of the visual axis in space is an important function because it determines the information that is provided by the visual system. Classically, the neural control of eye movements has been investigated in response to very simple and unrealistic visual stimuli (e.g. laser spots). In this condition, it is essentially the reflexive mechanisms that are brought to light.
In this thesis, we will investigate the predictive mechanisms that are used by the oculomotor system in response to more complex and realistic visual stimuli. In this condition, the eye movements are influenced by both the physical nature of the stimulus (size, color, speed) and associated cognitive cues (type of object, expected future trajectory, etc).
*****
The role of the visual feedback on the eyehand coordination during circular
arm movements with a handheld load in different gravitational fields
Olivier White, Philippe Lefèvre (CESAME) and J.L. Thonnard (READ)
UCLINMA
Abstract
In this experiment, we would like to study the eyehand coordination while subjects manipulate an object following a circular trajectory in novel gravity fields. The coupling between the grip force, normal to the grasped object, and the tangential force will be studied when the vectorial direction of the object’s acceleration is continuously changed in relation to the gravity during circular arm movements. When one moves an object in the frontal plane following a circular trajectory at a constant speed, the object is subjected to the vertical gravitational acceleration and to the centripetal acceleration. The tangential force tends to make the object slip out of the fingers. In order to restrain the object, the normal grip force has to be adequately adjusted to the tangential force fluctuations.
We are particularly interested in eye movements and the role played by visual feedback when performing this task in new gravity fields induced by parabolic flights (0G, 1G, 2G). We will record three sets of parameters. The parameters of movement dynamics (ATI 3D forces and torques transducer) will measure how the subject is able to anticipate the tangential force, which depends both on the gravity and the acceleration of the upper limb. The kinematics parameters measured with an OptoTrak 3020 system will indicate whether the subject is able to perform circular movements in altered gravitational environments with and without visual feedback. The eye movements measured with skalar system will report the strategy of the subject to follow the object’s trajectory.
*****
The development of statistical models for the analysis of microarray data
Joke Allemeersch, Bart De Moor & Yves Moreau
K.U.Leuven, Dept. Elektrotechniek, ESATSCD (SISTA), Heverlee
Abstract
Microarray technology allows us to analyze the activity of genes in biological samples on a massive scale. The analysis of such data requires rigorous statistical methods to deal with important problems like noise, reproducibility. The goal of this Ph.D. is to investigate the whole range of statistical techniques, like experimental design, power analysis, statistical testing, varianceanalysis, etc. on microarray data. In particular, a major goal is to extend current statistical techniques to the analysis of microarray timecourse experiments. Techniques such as dynamic programming, linear system modelling, and repeated measurement techniques will be investigated in this context.
*****
The application of advanced techniques for system identification to study side
channels for cryptographic algorithms: cryptanalysis and design
Evelyne Dewitte & Bart De Moor
K.U.Leuven, Dept. Elektrotechniek, ESATSCD (SISTA), Heverlee
Abstract
To protect information against active or passive eavesdropping, one uses cryptographic algorithms that have to offer a solid level of security. This security is threatened by a new kind of attacks, so called sidechannel attacks that use weaknesses in the implementation of the algorithms. We want to combine two previously unrelated subgroups, SISTA and COSIC to discover the limits of these attacks by using system identification and modelling tools on a MIMO model of the cryptographic system. Once these limits are known, one can consider countermeasures that are robust against the attacks.
*****
Highthroughput’ statistical analysis of microarray data
Steffen Durinck & Bart De Moor
K.U.Leuven, Dept. Elektrotechniek, ESATSCD (SISTA), Heverlee
Abstract
Microarrays form a powerful technique for functional genomics studies. We will develop methods and applications which will enable analysis of thousands of microarray experiments such as an advanced preprocessing step were a diagnosis tool is coupled with a systematic normalization procedure, methods to merge datasets from different origin, allow complex queries on the data, built a gene specific noise model and transcriptome maps.
*****
Applied Nonlinear Time Series Analysis
Marcelo Espinoza & Bart De Moor
K.U.Leuven, Dept. Elektrotechniek, ESATSCD (SISTA), Heverlee
Abstract
In the general context of modelling and forecasting of time series, important concepts coming from statistics and econometrics domains can be applied and further developed within the nonlinear modelling framework. This research will address these issues in order to develop practical models for the forecasting of electricity load and financial time series in general, in terms of input selection, data preprocessing, structure identification and time series dynamics.
*****
Contextdriven
mining of literature for intelligent knowledge management with largescale experiments
in the domain of functional genomics
Frizo Janssens, Bart De Moor & Yves Moreau
K.U.Leuven, Dept. Elektrotechniek, ESATSCD (SISTA), Heverlee
Abstract
To prolong the lifecycle of data coming from complex, expensive and large experiments in molecular biology, a knowledge management system with an intelligent data structure and querying possibilities will be developed. Contextdriven text mining techniques will enable automatic retrieval and management of information coming from heterogeneous sources on the Internet.
*****
Comparative study of the composition and evolution of bacterial regulons
Pieter Monsieurs & Prof. B. De Moor
K.U.Leuven, Dept. Elektrotechniek, ESATSCD (SISTA), Heverlee
Abstract
The composition of bacterial regulons differs significantly amongst bacterial species. During the first stage of the project the composition of bacterial regulons will be compared amongst enteropathogenic species. During a second stage we will try to identify the evolutionary processes that contributed to the detected variation in regulon composition. The developed methodology will be validated using two testsystems, the PhoPQ and the FNRregulon.
*****
Kernelbased Modelpredictive Control
Bert Pluymers, Bart De Moor & Johan Suykens
K.U.Leuven, Dept. Elektrotechniek, ESATSCD (SISTA), Heverlee
Abstract
Modelbased Predictive Control (MPC) is a control technique being used in wide sectors of industry in its linear form. When using nonlinear models however additional complications arise of which the most important is the fact that the involved optimization problems generally become nonconvex, which prohibits the online implementation of the controller. In this project will be examined in which way kernelbased methods can be used for constructing models of nonlinear systems for use in MPC, in which way methods of convex relaxation, used in classic neural network models, can be extended to these models and in which way robustness can be incorporated into these schemes.
*****
Microarray Data Analysis using Support Vector Machines and Kernel Methods
Nathalie Pochet, Bart De Moor & Johan Suykens
K.U.Leuven, Dept. Elektrotechniek, ESATSCD (SISTA), Heverlee
Abstract
The objective of this research is to find an optimal strategy for analyzing microarray data by using support vector machines and kernel methods. An important application area is oncology. Microarray data can be analyzed in three different ways: the discovery of diagnostic classes and genes with a similar behaviour, the performance of clinical and biological predictions, and the discovery of relevant genes and groups of genes.
*****
Model generation and reduction for acoustic phenomena
Bert Schiettecatte, Axel Nackaerts & Bart De Moor
K.U.Leuven, Dept. Elektrotechniek, ESATSCD (SISTA), Heverlee
Abstract
Physical sound models are being used more and more to calculate synthetic sound for realworld phenomena, for example, to experience the acoustic qualities of a room or to put a sound source in virtual space. The goal of this research project is to find a translation mechanism between a 3D description of a musical instrument or environment and a computationally efficient model. We hope to rely on psychoacoustics to reduce the complexity of both the 3D descriptions and the generated model.
*****
Development of a strategy based on phylogenetic footprinting for the identification
of regulatory elements in eukaryotic promoters
Ruth Van Hellemont, Bart De Moor & Y. Van de Peer
K.U.Leuven, Dept. Elektrotechniek, ESATSCD (SISTA), Heverlee
Abstract
The goal of this research is the development of a generic procedure for phylogenetic footprinting for identification of regulatory motifs in eukaryotic sequences. This will consist of two subgoals: first, a methodology will be developed for the generation of a data set appropriate for phylogenetic footprinting. In a second phase, the methodology for phylogenetic footprinting will be optimised. The developed methodology will be validated by a test set of hox genes.
*****
Design and evaluation of DSP algorithms for feedback cancellation in Public
Address systems
(Ontwerp en evaluatie van digitale signaalverwerkingsalgoritmen voor feedbackonderdrukking
in 'Public Address'systemen)
Toon van Waterschoot & Marc Moonen
K.U.Leuven, Dept. Elektrotechniek, ESATSCD (SISTA), Heverlee
Abstract
In a Public Address (P.A.) system the loudspeaker sound is often fed back into the microphone. This may result in system instability that is perceived as “howling”. Stateoftheart solutions use notch filters in the forward path between microphone and loudspeaker. Our research will focus on another approach, inspired by a similar technique in hearing aids. This approach aims at feedback cancellation by estimating the feedback signal using an adaptive filter.
*****
Internal Normalisation in LAICPMS & NonLinear Growth Rates in Biota
Fjo De Ridder(1), Anouk Verheyden(2), Phillippe Willenz(3), Frank Dehairs(4),
Johan Schoukens(1), Rik Pintelon(1)
(1) Department of Electricity and Instrumentation, Vrije Universiteit Brussel,
Pleinlaan, 2, 1050 Brussels, Belgium, federid@pop.vub.ac.be
(2) Laboratory of General Botany and Nature Management, Vrije Universteit Brussel
(3) Department of Invertebrates, Royal Belgian Institute of Natural Sciences,
Rue Vautier 29, 1000 Brussels, Belgium
(4) Analytical and Environmental Chemistry Department, Vrije Universteit Brussel
Abstract
1. Due to fluctuations in the laser intensity and differences in the density of the sampled material, the signal of an LAICPMS instrument is not proportional to the concentration of a specific element. This effect is known as drift and is usually corrected by referring the analyte signal to the one of an internal standard. The latter is an element homogeneously distributed throughout the whole sample. Next, the drift pattern will be reflected in the measured pattern for this element and the signals of all other elements can be compensated for this estimated drift pattern. When this strategy was employed on measurements of a sclerosponge (a calcium carbonate secreting marine sponge), it was found that different internal standards (Ca, Sr, U, …) led to different drift patterns. We developed a weighted least square estimation of the drift pattern, based on the measurements of multiple internal standards. Weighing emphasises the more precise measurements and thus estimates the most probable drift pattern from all internal standards at once. The use of more internal standards enabled us to have an internal quality check.
2. For specific biota the record of a feature (i.e. a proxy) along a growth axis can reflect (changing) environmental conditions experienced during lifetime of the organisms. For example, the density of vessels in trees, are related to ambient environmental conditions. For a mangrove tree it was attempted to partly reconstruct these environmental conditions by analyzing vessel density along the growth axis. When assuming linear growth rate (in this case 1.3mm/year) this led to a reasonable error in the matching of environmental conditions with vessel density. The second part of this poster describes an estimation of this nonlinear growth rate based on methods used to characterize time base distortions in high frequency sampling scopes.
*****
PowellSabin splines
Jan Maes
KULNUMERICS
Abstract
Piecewise polynomials on triangulations, and PowellSabin (PS) splines in particular, form an attractive alternative for the widely used tensor product splines.
In this poster we give an overview of the research on PowellSabin spline surfaces. We present a normalised Bspline representation for PS splines. With this representation we can form a set of control triangles which have a nice geometric interpretation. A subdivision scheme for PS splines is introduced both on uniform and general triangulations. New wavelets have been generated by using these subdivision schemes as the prediction step in the lifting scheme. We also show some algorithms for data fitting and surface interrogation. Finally we give some suggestions for further research.
*****
DDEBIFTOOL, a software package for the bifurcation analysis of Delay Differential
Equations
G. Samaey, D. Roose, K. Engelborghs, T. Luzyanina
KULNUMERICS
Abstract
DDEBIFTOOL is a Matlabbased software package for numerical bifurcation analysis of delay differential equations with fixed and/or statedependent delays. The package contains procedures for stability analysis of steady state solutions of DDEs, computation of periodic solutions and their stability (using a collocation approach) and computation of homoclinic and heteroclinic orbits.
We illustrate the capabilities of DDEBIFTOOL for the analysis of mathematical model from variuous application areas.
*****
Macroscopic analysis of microscopic evolution laws
Pieter Van Leemput
KULNUMERICS
Abstract
Looking for models that capture the complexity of dynamical systems, the microscopic Lattice Boltzmann model can be considered as a viable alternative to the macroscopic description of the system by means of a set of partial differential equations. A macroscopic time integrator that uses the microscopic Lattice Boltzmann evolution law instead of a macroscopic one, is constructed. Our ultimate goal is the derivation of the important macroscopic properties of the dynamical system using this time integrator. For example, the stability of the system can be analyzed accurately by computing the largest eigenvalues from the spectrum of the macroscopic time integrator. Also, the macroscopic time integrator can be coupled to the NewtonPicard method to allow for the continuation of solution branches and a bifurcation analysis of the system. The one dimensional FitzHughNagumo reactiondiffusion system is considered as an example. Other microscopic systems, like cellular automata, will be considered as well.
*****
Numerical bifurcation analysis of largescale delay differential equations
Koen Verheyden (1)
Scientific Computing Research Group, Department of Computer Science, K.U.Leuven.ULgSYST
Abstract
This Ph.D. thesis is about the bifurcation analysis of largescale delay differential equations (DDEs). More precisely, the design and the implementation of efficient and robust iterative methods.
Delay differential equations are used more and more as a modelling tool in, e.g., control theory and population dynamics. The delay terms take “memory phenomena” such as latency into account. Fixed point delay models as well as continuous delay models are being used. The numerical treatment of both cases will be considered in this thesis. For the continuous delay case – where the delay is given in the form of an integral term – adapted discretization techniques will be studied. Another key issue is the use of appropriate iterative numerical linear algebra techniques.
We have already implemented an iterative method for the computation of periodic solutions of DDEs. The initial value problem for a DDE requires the function to be given on an entire interval. Thus the computation of periodic solutions always results in large nonlinear systems because the periodicity condition has to be discretized on an initial and a final interval.
We use single shooting on the linearization of the DDE about a proposed trajectory. The time integrator we use is the implicit RungeKutta scheme corresponding to the GaussLegendre collocation method. One linear single shooting problem – corresponding to a Newton step – is solved by iterative refinement. The resulting linear systems are approximately solved by the NewtonPicard method. This iterative method exploits the fact that there are only a few semistable or unstable Floquet multipliers.
Our implementation
is done in DDEBIFTOOL, a Matlab package for the numerical bifurcation analysis
of DDEs developed at the Scientific Computing Research Group.
(1) Research Assistant of the Fund for Scientific Research  Flanders (Belgium).
*****
Experimental Study and Modelling of the Neural Control of Juggling
Renaud Ronsse
Montefiore Department  Université de Liège (ULg)
Abstract
This research aims at a better understanding of the key control parameters that enable a human to perform simple juggling tasks. Concurrent research work is studying the dynamics and mathematical control of this task (see M.Gerard, ULg). This project will confront theoretical predictions based on some conclusions given by mathematical results to an experimental study of human juggling. Juggling is here understood in the broad sense of a task that is inherently unstable, thus requiring some amount of feedback, and that involves rhythmic coordination. For a human juggler, feedback can be given by a visual or a tactile information.
*****
Modeling the competitive growth of Listeria innocua and Lactococcus lactis
M. Antwi, K.M. Vereecken and J.F.M. Van Impe
BioTeCBioprocess Technology and Control, Katholieke Universiteit Leuven
Kasteelpark Arenberg 22, B3001 Belgium
Abstract
The knowlegde of antagonistic interaction phenomena in mixed microbial cultures is essential for microbial safety and shelf life estimation of food products. In this work, the growth of Listeria innocua and Lactococcus lactis in a modified Brain Heart Infusion medium was studied in pure and mixed cultures. An existing single species model is used to describe the experimental data. The model provides a reasonable description of the mixed population growth studied but the validity region may be limited. In further research, a novel model that includes the factors affecting the microorganisms' metabolism will be applied.
*****
Modeling the onset of filamentous bulking based on image analysis information
in wastewater treatment systems
E.N. Banadda, R. Jenne, I. Smets and J.F. Van Impe
BioTeCBioprocess Technology and Control, Katholieke Universiteit Leuven
Kasteelpark Arenberg 22, B3001 Belgium
Abstract
Filamentous bulking is a problem widespread in the operation of activated sludge process. A fully automated image analysis method for recognizing and characterizing flocs and filaments in activated sludge images has been developed. It is the aim of this work to seek correlations between image analysis information with classical measurements and to investigate whether image analysis information can be used to predict the onset of filamentous bulking with aid of black box models.
*****
Application of a novel procedure to quantify thermal inactivation kinetics (Part
II)
Valdramidis V.P., Bernaerts K., Geeraerd A.H. and Van Impe J.F*.
BioTeCBioprocess Technology and Control, Katholieke Universiteit Leuven
Kasteelpark Arenberg 22, B3001 Belgium
Abstract
The concept of predictive microbiology is that a detailed knowledge of the responses of microorganisms to environmental conditions enables objective evaluation of effects of processing, distribution and storage operations on the microbiological safety and quality of food. Kinetic parameters and models are the main tools provided for the implementation of predictive microbiology for the different preservation processes. Among these processes, thermal treatment is a preservation process that has been practiced for more than five thousands years and is undoubtedly the method most widely used in food industry to inactivate microorganisms. Inactivation kinetics is of high interest for the determination of the amount of microbial inactivation.
This paper deals with estimation of thermal microbial inactivation kinetics using the methodology of optimal experiment design and the advanced processing technique as presented in Bernaerts and Van Impe (2002) [Food Micro 2002]. Time and temperature parameters are estimated, namely, the decimal reduction time (Dvalue) and the thermal resistance constant (zvalue). Integrated into the Bigelow model they can yield predictions for the specific inactivation rate at certain temperatures.
E.coli K12, grown in Brain Heart Infusion, is chosen as a surrogate for the food borne pathogen E.coli O157:H7. Survival data of the microorganism are collected and processed according to the optimal experiment design. The population density data are described accurately by a dynamic inactivation model combined with the Bigelow model through nonlinear regression. The parameter estimation accuracy on D and z is assessed by construction of joint confidence regions.
The magnitude of the estimated parameters is in a relevant range with previously published data. Furthermore, the parameters were estimated by much less experimental effort while in literature, a higher number of temperature series is applied. Finally, the joint confidence region of the bestfit parameter estimates when fitting the processed data is highly satisfactory.
Acknowledgements: This research is supported by the Research Council of the Katholieke Universiteit Leuven as part of projects OT/99/24 and IDO/00/008, the Institute for the Promotion of Innovation by Science and Technology (IWT), the Fund for Scientific Research – Flanders (FWO) as part of project G.0213.02, the Belgian Program on Interuniversity Poles of Attraction and the Second Multiannual Scientific Support Plan for a Sustainable Development Policy, initiated by the Belgian State, Prime Minister’s Office for Science, Technology and Culture, and the European Commission as part of project EU QLK1CT200101415. The scientific responsibility is assumed by its authors.
Poster Session II
LQOptimal Temperature Regulation for Nonisothermal Plug Flow Reactor
I. Aksikas, J.J. Winkin, D. Dochain
UCLINMA
Abstract
The LinearQuadratic optimal temperature regulation problem is studied for a nonisothermal plug flow tubular reactor model. The problem is solved for the linearized model around a constant temperature profile along the reactor.Then the resulting state feedback is applied to the nonlinear model, and the corresponding closedloop system performances are analyzed.
*****
The Time Course of Compensation for Anticipatory Smooth Eye Movements in a Target
Localization Task
G. Blohm, M. Crommelinck, M. Missal and P. Lefèvre
CESAME and Lab. Neurophysiol., Université catholique de Louvain, Belgium
Abstract
A target briefly flashed during smooth pursuit eye movements evokes localization saccades that do not compensate for the ongoing smooth eye movement (McKenzie and Lisberger 1986). In this study, we use a flash localization task to investigate the compensation mechanism of anticipatory smooth eye movements.
After a fixation (800ms) and a gap period (300ms), 7 human subjects were required to pursue a moving target (40°/s). Repetitive presentation of this stimulus led to robust anticipatory smooth eye movements (10°/s). In 30% of the trials, instead of the ramp, a peripheral 10ms flash appeared ±15° around the current eye position. Subjects were instructed to make saccades to the remembered location of the flash in darkness.
We found that, on average, the smooth eye movement lasted until 540ms after the flash. The capture of the target typically required 2 to 4 saccades because the smooth eye displacement (SED) perturbed the orientation to the flash. The compensation of this perturbation was a dynamical process that, on average, started only ~300ms after the flash. At the end of the orientation process (on average 849ms after the flash), the oculomotor system compensated for 70% of the SED (43%92% across subjects).
We conclude that the compensation of the SED must be based on an efference copy of the smooth anticipatory motor command that updates the spatial localization of the flashed target. In addition, there is a 300ms delay in this compensation that includes the time necessary to program and execute saccades but also for the integration of anticipatory motion.
Supported by FNRS, SSTC and FSR (Belgium).
*****
Observability analysis of a nonlinear tubular bioreactor
Cédric Delattre
UCLINMA
Abstract
In this poster an observability analysis is performed for an axial dispersion tubular bioreactor. It involves one growth reaction that follows the nonlinear Monod law.
As a first step, it is assumed that the biomass is constant. In this case, the process can be described by a semilinear parabolic Partial Differential Equation (PDE). More precisely, the analysis is performed on a tangent linearized model, that is described by a linear PDE with a spatialdependent coefficient. It is reported that the associated linear infinitedimensional operator is a SturmLiouville operator, a Rieszspectral operator and generates a $C_{0}$semigroup. Then it is shown that a finite number of dominant modes of the system are observable when the substrate concentration is measured at the reactor output by an appropriate sensor.
Open problem: it is now assumed that the biomass varies. How to analyze the observability of this new model? Will both studies give consistent results?
*****
Switched Continuous (Hybrid) Model of a High Pressure Food Thawing
Hadyan Fibrianto
UCLINMA
Abstract
The refrigeration of foodstuffs is an important step in food industries. Recent studies have established that the kinetics of freezing and thawing processes is crucial in the final quality of the food [Chourot, 1997][Chevalier et al., 2000]. Therefore, it is important to master this kinetics. High pressure makes the control of such a process easier by modifying the thermophysical properties of water. That’s why a high pressure food thawing is considered here.
Indeed, such a process involves a complex modelling: distributed parameters and nonlinear. This nonlinearity is due to firstly, phase changes (solid, melting, liquid) and secondly, pressure steps.
This paper shows that hybrid methodology allows to linearize the model. Some laboratory experiments are carried out to validate the model.
*****
Selfadjoint quadratic eigenvalue problems and associated distance problems
Yvan Hachez
UCLINMA
Abstract
Two important classes of quadratic eigenvalue problems are composed by elliptic and hyperbolic problems. In N. J. Higham, F. Tisseur, and P. M. Van Dooren, Linear Algebra Appl., 351352 (2002), pp. 455  474, the distance to the nearest nonhyperbolic or nonelliptic quadratic eigenvalue problem is obtained using a global minimization problem. This paper proposes explicit formulas to compute these distances and the optimal perturbations. The problem of computing the nearest elliptic or hyperbolic quadratic eigenvalue problem is also solved. Numerical results are given to illustrate the theory.
*****
Calculation of elementary flux modes in reaction networks: illustration with
CHO cell metabolism
Agnès Provost
UCLINMA
Abstract
When considering a complex reaction network, an interesting question is to characterize the simplest reactions that can connect input and output species. In this poster, we present a method based on elementary flux modes (EFM) which are obtained by computing the convex basis of the kernel of the stoichiometric matrix. An algorithm is presented for the calculation of EFM. The successive steps of the method are illustrated on the network describing CHO cells central metabolism.
*****
Catchup Saccades in Two Dimensions
C. Schreiber, G. Blohm, M. Missal & P. Lefèvre
CESAME and Lab. Neurophysiol., UCL, Brussels, Belgium
Abstract
During visual tracking of a moving stimulus, primates orient their visual axis by combining smooth pursuit and catchup saccades. A quantitative analysis of catchup saccades has been done for horizontal movements (de Brouwer et al., 2002). In this study, we investigate the properties of catchup saccades made to visual stimuli moving in two dimensions (2D).
We measured 2D eye movements in 4 human subjects (search coil technique). Each trial started with a 2D Rashbass stepramp stimulus (velocity [10..20 deg/s], direction [0..360 deg], duration [600..1100 ms]). This was followed by a second stepramp of the target [500..700 ms]. Both the direction of the position step (PS, [10..10 deg]) and the velocity step (VS, [40..40 deg/s]) varied randomly in 2D. We analysed the first catchup saccades after the second step of the target.
We found that catchup saccades to the second ramp were characterized by latencies as short as 100 ms with respect to the second target step (mean 185 ms). The average gain of catchup saccades was 0.84 (n=1175). However, when data were restricted to moderate values of retinal slip (RS<20 deg/s), the average gain was 0.9 (n=569). A multiple linear regression analysis was performed to find the parameters determining the amplitude of catchup saccades. For both horizontal and vertical components of saccades, we found the best correlation with position error (PE) and RS as independent variables: Amp=0.9*PE+0.11*RS (R>0.97, n=569).
We conclude that both position error and retinal slip are taken into account for programming 2D catchup saccades.
Supported by FNRS, SSTC and FSR (Belgium).
*****
Optimal prefiltering in Iterative Feedback Tuning
Gabriel Solari
UCLINMA
Abstract
IFT is a method to tune a controller performing special experiments on the plant with periods where the operating conditions are normal. We present a method to improve the convergence of IFT by prefiltering the input data used in the special experiment. The optimal prefilter is computed at every iteration from data collected under normal operating conditions of the plant.
*****
TOUCAN: Deciphering the CisRegulatory Logic of Coregulated Genes
Stein Aerts, Gert Thijs, Bert Coessens, Mik Staes, Yves Moreau, & Bart De
Moor
K.U.Leuven, Dept. Elektrotechniek, ESATSCD (SISTA), Heverlee
Abstract
Toucan is a Java application for the rapid discovery of significant cisregulatory elements from sets of coexpressed or coregulated genes. Biologists can automatically (1) retrieve genes and intergenic regions, (2) identify putative regulatory regions, (3) score sequences for known transcription factor binding sites, (4) identify candidate motifs for unknown binding sites, and (5) detect those statistically overrepresented sites that are characteristic for a gene set.
*****
Integration of Datamining techniques in a CPI environment (BASF)
Steven Bex & Bart De Moor
K.U.Leuven, Dept. Elektrotechniek, ESATSCD (SISTA), Heverlee
Abstract
ESATSCD has an extensive collection of datamining algorithms, preprocessing techniques, and other modelling and analysis tools to go from large data sets to useable knowledge. However, most of these are either for very specific applications, or are centered in the Bioinformatics domain. BASF on the other hand, has experience with datamining for specific problems, but does not use the newest algorithms and cannot assess which algorithm is best used where. Hence this project, in which a feasibility study/integration will be done of SCDtechniques in a BASF environment.
*****
Building a Semantic BioScape
Bert Coessens, Janick Mathijs and Bart De Moor
K.U.Leuven, Dept. Elektrotechniek, ESATSCD (SISTA), Heverlee
Abstract
Biology is a knowledgebased rather than an axiombased science. As a consequence, new knowledge has to be inferred from and validated by existing knowledge. But the huge amounts of existing knowledge, it's heterogeneous nature and the fact that it is distributed all over the internet, make this a horrendous task. Biology copes with a problem of diversity and distribution of the information it works with.
We propose an architecture that integrates different tools and algorithms for analysis and mining of microarray data and gives them integrated access to heterogeneous information sources.
The general concept is not to integrate all the different tools, algorithms and biological information sources in the architecture, but rather construct an integrated view of the information they manage, output and contain in the form of applicationbased ontologies.
The SOAP Web Service technology and DAMLS are used to make different distributed algorithms searchable and interoperable and to hide their specific implementations from the architecture.
With this architecture we want to make analysis of microarray data faster and more efficient.
*****
The canonical decomposition: Uniqueness, Algorithm, Applications
Lieven De Lathauwer, Bart De Moor and Joos Vandewalle
K.U.Leuven, Dept. Elektrotechniek, ESATSCD (SISTA), Heverlee
Abstract
The Canonical Decomposition (CD) of a higherorder tensor is the decomposition of that tensor in a minimal sum of rank1 terms. Unlike the situation for matrices, this decomposition can be unique without orthogonality constraints on the components, and the minimal number of terms (the "tensor rank") can be much bigger than the dimensions of the tensor.
In this poster we present a new uniqueness theorem for the CD, which applies to a wide range of tensors for which uniqueness had not been proven yet.
The theorem is constructive; it shows that the components may be obtained from a simultaneous congruence transformation. Applications can be found in factor analysis, harmonic retrieval, underdetermined independent component analysis, blind identification, stateoftheart DSCDMA techniques, etc.
*****
Using Literature and Data to Annotate and Learn Bayesian Networks
Geert Fannes, Patrick Glenisson & Bart De Moor
K.U.Leuven, Dept. Elektrotechniek, ESATSCD (SISTA), Heverlee
Abstract
The increasing availability of electronic literature opens up the possibility of using it as prior knowledge when dealing with complex statistical models where data is scarce or high levels of noise are present. This raises the question of how to actually perform the integration of domain literature with statistical data. In this paper, we assume that our textual information consists of short freetext descriptions of the domain variables and, optionally, of a large repository of related domain literature while the statistical data are classical vectors of observations. Because of their explicit semantics and powerful computational methods, Bayesian networks are suitable candidates for the integration of prior domain knowledge and statistical data. To connect our computational model with textual domain knowledge, we define an extended representation of Bayesian networks called the Annotated Bayesian Network. We introduce a textbased prior for the evaluation of Bayesian network substructures. We derive a textbased prior distribution over the space of Bayesian network structures and update this to a posterior with data. We evaluate our methodology in the ovarian cancer domain by first indicating the correlation between the textbased and databased scores for substructures. Next, we compare the performance of our newly introduced textbased prior for Bayesian network structures in classifying ovarian tumors versus the performance of a Bayesian network with a uniform structure prior. We show that in the case where only a small number of samples is available (which is a common situation in medical and biological applications), the prior derived from the literature increases the classification performance.
*****
Soccer Mining: Analysis and Design
Jelle Geerits, Emil Muresan & Bart De Moor
K.U.Leuven, Dept. Elektrotechniek, ESATSCD (SISTA), Heverlee
Abstract
This poster presents the ESATSCD soccer analysis and data mining project. Starting from four videos of a soccer game, taped with static cameras from four different corners, the 2D coordinates of the players on the field and the 3D coordinates of the ball are retrieved and stored in a database.
Starting from this database, statistics and analyses can be performed. Trivial statistics, like average speed, acceleration, total run distance and field coverage of each player, are visualised. Also more complex analyses are explored, like: detection of game strategy, prediction of best actions in a given situation, offside detection, referee positioning with respect to the game, modelling power consumption, correlation diagrams with physiological measurements such as heart beats, etc.
In addition, we work on automated label generation (closetooffside, dangerforcounterattack, bestpassopportunity, etc) to be integrated in VideoCoach, another tool that we have developed.
This project turns out to be truly multidisciplinary: Player/ball tracking via Kalman and/or particle filtering, computational geometry (player polygons), ball track fitting (orthogonal distance regression, curvature and torsion (FrenetSerret) coordinates), optimal control theory (MPC for optimal defense determination) and game theory (best attack under uncertainty). Of course, we'll show some game sequences.
*****
Subspace regressionin reproducing kernel Hilbert space
L. Hoegaerts, J.A.K. Suykens, J. Vandewalle & B. De Moor
K.U.Leuven, Dept. Elektrotechniek, ESATSCD (SISTA), Heverlee
Abstract
We focus on three methods for finding a suitable subspace for regression in a reproducing kernel Hilbert space:
* kernel principal
component analysis, kernel partial least squares and kernel canonical correlation
analysis, and
* we demonstrate how this fits within a more general context of subspace regression.
For the kernel partial least squares case a least squares support vector machine style derivation is given with a primaldual optimization problem formulation.
The methods are illustrated and compared on a number of examples.
*****
A Case Study on Traffic Flow Modelling, Simulation and Control
Sven Maerivoet, Tom Bellemans and Bart De Moor
Department of Electrical Engineering, ESATSCD (SISTA), Katholieke Universiteit
Leuven
Abstract
In our case study, we observe the traffic data (as collected by the Traffic Centre in Wilrijk) of the E17 highway in the direction of Antwerp. Interpretation of this timeseries leads to the wellknown fundamental diagrams from traffic flow theory (these diagrams exhibit the metastability and hysteresis phenomena). As a macroscopic traffic flow model, we use Papageorgiou's METANET model. The microscopic model is in one case based on a continuous implementation and in another case on a discrete version implemented as a traffic cellular automaton. As a control measure, we investigate the use of ramp metering using model predictive control, which outperforms the default ALINEAalgorithm.
*****
Robust crossvalidation score functions for kernel based function estimation
Kristiaan Pelckmans, Jos De Brabanter, Johan Suykens & Bart De Moor
K.U.Leuven, Dept. Elektrotechniek, ESATSCD (SISTA), Heverlee
Abstract
This research focuses on methods for tuning of the hyperparameters for function estimation in the case of nonGaussian noise or outliers on the output data. Although the classical techniques such as leaveoneout and generalized crossvalidation are widely used, they are not able to handle certain classes of contaminated noise models effectively. For this purpose, a robustified version of the Lfold crossvalidation and generalized crossvalidation are introduced. The problem of optimizing the robust scorefunction is considered. As empirical evaluations of the scorefunction on a number of testexamples illustrate, the resulting scorefunction may suffer from the existence of several local minima. These are mainly caused by variance on the score estimator. A suitable optimization procedure is developed then based on a stochastic approximation method leading to the global minimum of the score function. The ideas are illustrated for weighted LSSVMs with RBF kernel. The robust crossvalidation methods are applied to several toy and real life data sets with improved test set performance measured in different norms.
*****
A family of entanglement monotones for mixed 2bit quantum systems
Maarten Van den Nest & Bart De Moor
K.U.Leuven, Dept. Elektrotechniek, ESATSCD (SISTA), Heverlee
Abstract
We present an infinite class of entanglement monotones for general 2bit quantum systems (qubits), using the Lorentz singular value decomposition of a 2qubit density matrix. These entanglement monotones are linear functions of the Lorentz singular values and comprise a generalization of the concurrence, a celebrated entanglement measure for 2qubit systems. Our family of monotones provides necessary conditions for one state to be convertible to another using local quantum operations and classical communication.
*****
Intelligent Information Infrastructure supporting Knowledge Management in a
Research and Development Environment
Dries Van Dromme, & Bart De Moor
K.U.Leuven, Dept. Elektrotechniek, ESATSCD (SISTA), Heverlee
Abstract
In the project “McKnow”, we set up an information system that considers both user profiles and documents (of whatever type, and their metadata) as resources. In our research, it is our intention to also capture relevance feedback, information from interactions of users with documents, and socalled “tacit knowledge” from experts. All these data feed the Data Warehouse, which, coupled with good Data Mining techniques, will yield an automated, useroriented, dynamic system for knowledge management in a chaotic environment.
In particular, using clustering algorithms, we've been grouping the documents from ESATsista publications to see if the clusters reflect the Research Group's Divisions in both scientific topics and composition by its members (= users = researchers). To be able to do this, we had to develop an automated text extraction and indexing method. Considerable effort has gone into these Text Mining aspects, and we will briefly discuss them.
Next, the results of the Clustering efforts will be discussed and shown.
*****
Identification of Young's Modulus from Broadband Modal Analysis Experiments
R. Pintelon(1), P. Guillaume(2), K. De Belder(1) and Y. Rolain(1)
(1)Vrije Universiteit Brussel, dept. ELEC, Pleinlaan 2, 1050 Brussels, BELGIUM
(2)Vrije Universiteit Brussel, dept. WERK, Pleinlaan 2, 1050 Brussels, BELGIUM
Abstract
The stressstrain relationship of linear viscoelastic materials is characterized by a complexvalued, frequency dependent elastic modulus (Young's modulus). Using system identification techniques it is shown in this paper how can be measured accurately in a broad frequency band from forced flexural (transverse) and longitudinal vibration experiments on a beam under freefree boundary conditions. The advantages of the proposed method are (i) it takes into account the disturbing noise and the nonlinear distortions, (ii) is delivered with an uncertainty bound, (iii) the low sensitivity to nonidealities of the experimental set up, and (iv) the ability to measure lowly damped materials. The approach is illustrated on several practical examples: brass, cupper, plexiglass and PVC.
*****
New results concerning the new modified Newton iteration for Toeplitz matrices
Gianni Codevico
KULNUMERICS
Abstract
The classical Newton iteration method for matrices can be modified into an efficient algorithm when structured matrices are involved. The difficulty, however, is the importance of the choice of the starting matrix and the compressed form of the inverse approximation. With this poster, we want to show our recent progresses on the new modified Newton iteration. The validity of the approach is illustrated by numerical experiments based on our C++ library for Toeplitlike systems.
*****
Divide and conquer algorithms for computing the eigendecomposition of diagonalplussemiseparable
matrices
Ellen Van Camp
KULNUMERICS
Abstract
First we construct a new algorithm with quadratic computational complexity to tridiagonalise a symmetric diagonalplussemiseparable matrix. From this algorithm we derive two divide and conquer algorithms (a oneway and a twoway algorithm) for calculating the eigendecomposition of a symmetric diagonal plus semiseparable matrix.
Keywords: diagonalplussemiseparable matrices, eigendecomposition, divide and conquer algorithms, fast and stable algorithms.
*****
An algorithm for computing the singular values based on semiseparable matrices
Raf Vandebril, Marc Van Barel & Nicola Mastronardi
KULNUMERICS
Abstract
The poster presents a new algorithm for finding the singular values of an m by n matrix A. The traditional algorithm based on bidiagonal matrices and the implicit application of the QR algorithm are explained. The new algorithm is based on semiseparable matrices instead of bidiagonal ones. In every step a comparison is made between the traditional and this new approach. The traditional reduction to an upper bidiagonal is replaced with a reduction to an upper semiseparable matrix. The implicit QR algorithm for the bidiagonal matrices is translated towards the semiseparable case. Numerical experiments are shown, comparing the accuracy and the timings of the traditional algorithm with the new algorithm.
*****
Stabilization of Planar Juggling Patterns
Manuel Gérard
ULgSYST
Abstract
This poster is devoted to the stabilization of planar juggling trajectories. A juggling pattern is defined by the periodic motion of a masspoint bouncing between the edges of a rectangular billiard. The model assumes Newtonian free flight dynamics and elastic collisions. The angular position and angular momentum of one of the edges are actuated. The stabilization problem is analyzed by studying the discrete return map of the ball on this edge. When the system is uncontrolled, two physical quantities are shown to be conserved at the instants of impact: the energy of the ball and its velocity component orthogonal to the surface. Based on these invariant quantities, discrete control laws are designed to stabilize a specified juggling figure. It is argued this control methodology achieves large basins of attraction, simplifies the closedloop analysis and applies to a wide range of juggling configurations.
*****
Image analysis as a monitoring tool for activated sludge properties
R. Jenné, E.N. Banadda and J.F. Van Impe
BioTeC  Bioprocess Technology and Control, Katholieke Universiteit Leuven
Kasteelpark Arenberg 22, B3001 Leuven, Belgium
Abstract
A filamentous bulking problem occurs when filamentous bacteria overgrow (wellsettling) floc forming bacteria. In this work, a fully automatic image analysis procedure was developed to monitor the activated sludge properties in a lab scale installation. In this way, the potential of image analysis as an early warning system for filamentous bulking was evaluated.
*****
Towards a new generation of simple models for microbial growth
Poschet F., Vereecken K.M., Geeraerd A.H. and Van Impe J.F.*
BioTeC  Bioprocess Technology and Control, Katholieke Universiteit Leuven
Kasteelpark Arenberg 22, B3001 Leuven, Belgium
Abstract
Food safety and quality are largely determined by the proliferation of pathogenic and spoilage microorganisms during the life cycle of the product (i.e., from the start of the production process until consumption). In order to simulate and predict the microbial evolution in foods, mathematical models are developed in the field of predictive microbiology.
Microbial growth normally passes 3 main periods: first a lag phase during which the cells adapt to their new environment, followed by an exponential growth phase during which the cells multiply exponentially, and finally a stationary phase during which a maximum population density is reached. The growth model of Baranyi and Roberts, which is considered to be the most widespread growth model, describes these 3 phases in microbial growth by means of the following 2 differential equations:
N [CFU/mL] is the microbial cell concentration, mmax [1/h] the maximum specific growth rate and Q [] characterises the physiological state of the microbial cells. The factor Q/(1+Q) is called the adjustment function and describes the lag phase by means of the physiological state of the cells. The factor mmaxN describes the exponential growth phase. The (1N/Nmax) factor, the inhibition function, describes the transition to the stationary phase. A major disadvantage of the Baranyi and Roberts growth model is the lack of mechanistic knowledge in the inhibition function and its poor extendibility to interactions between different microbial species, microbial growth in structured media, etc.
In our new modelling concept, the inhibition function of the Baranyi and Roberts growth model is replaced by a more mechanistically based function. The stationary phase is assumed to be a result of an increasing inhibition by a toxic metabolite produced by the cells.
After a careful structural analysis, the new model is applied to a large set of experimental data of Escherichia coli K12. Both aspects are compared with the growth model of Baranyi and Roberts. The new model has a more mechanistically founded basis as compared to the model of Baranyi and Roberts and is by consequence easier to extend to more realistic (and more complex) situations (e.g., microbial interactions and growth in a structured environment), while maintaining similar predictive value.
Acknowledgements: This research is supported by the Research Council of the Katholieke Universiteit Leuven as part of projects OT/99/24 and IDO/00/008, the Institute for the Promotion of Innovation by Science and Technology (IWT), the Fund for Scientific Research–Flanders (FWO) as part of project G.0213.02, the Belgian Program on Interuniversity Poles of Attraction and the Second Multiannual Scientific Support Plan for a Sustainable Development Policy, initiated by the Belgian State, Prime Minister’s Office for Science, Technology and Culture, and the European Commission as part of project QLK1CT200101415. The scientific responsibility is assumed by its authors.
*****
Analysis and evaluation of a serial dilution experimental protocol by means
of a simulation model
A.R. Standaert, A.H. Geeraerd, K. Bernaerts, K. Francois, F. Devlieghere, J.
Debevere and J.F. Van Impe
BioTeC  Bioprocess Technology and Control, Katholieke Universiteit Leuven
Kasteelpark Arenberg 22, B3001 Leuven, Belgium
Abstract
Serial dilution experiments are a standard way of obtaining single cells for study of, in this case, individual cell lag times. A simulation model for the serial dilution experimental protocol is constructed providing further insights in the experimental process, and is used to suggest improvements of the experimental protocol.
Addenda
Learning
in Markov models using the imprecise Dirichlet model
Erik Quaeghebeur
RUGSYSTeMS
Abstract
The objective of our research is first of all the development of a method for learning the transition probabilities in (possibly hidden) Markov models using imprecise probabilities, and next the application of this method to some reallife problems. The learning model used will be the imprecise Dirichlet model, an extension of the precise Dirichlet model to the theory of imprecise probabilities. Possible applications are genesequence alignment and prefetching of web pages.