[Upcoming events]  [PhD defenses overview]  SMC events archive
Wed 15 - Wed 15 May-13 WG2/WG3 Seminar - Marco Reis
CIT 00.01
2:00 pm-3:00 pm

"Image-based process monitoring: towards an integrated framework"

Prof. Marco Reis
(Department of Chemical Engineering, University of Coimbra, Portugal)

Image Analysis procedures have been increasingly applied in industry for handling problems where the quality of products depends upon characteristics that can be captured or inferred by means of image sensing devices. From our perspective, images are a particular case of a general class of data structures called “profiles”. In such an abstract description, a time-series of a process variable or the accurate surface contour of a mechanical piece measured through profilometry constitute examples of 1D profiles; a grey-level image or a time-course cDNA microarray experiment do correspond to 2D profiles; multispectral images or multivariate batch data are examples of 3D profiles (and the sequence can proceed to higher dimensions, for instance with measurements from hyphenated instruments). In this talk, we propose a transversal look to the variety of “profiles” we are currently handling in industry and propose frameworks for handling them in an integrated, coherent and scalable way, regarding the number of dimensions or modes in the profile.

slides

Mon 13 - Mon 13 May-13 Quarterly Meeting of OPTEC WG3
CIT 01.06
2:00 pm-4:00 pm
"Quarterly Meeting of OPTEC Working Group 3 on Parameter and State Estimation"

The next WG3 meeting will focus on online estimation with presentations
by Milan Vukov (Moving Horizon Estimation) and Rien Quirynen (dedicated integration tools).

The final schedule will be given later.

Thu 2 - Thu 2 May-13 SISTA Seminar - Diego Peluffo
ESAT 00.62
3:00 pm

"Kernel Spectral Clustering for dynamic data"

Diego Peluffo (Universidad Nacional de Colombia - Manizales)

Spectral clustering has taken an important place in pattern recognition due to its capability of accurately grouping data having complex structure. There are several spectral clustering approaches mainly related to graph partitioning. The most suitable techniques are those based on kernels.
Nevertheless, one of the biggest disadvantages of spectral clustering techniques is that most of them have been designed for static data analysis, that is to say, without taking into consideration the changes along the time (i.e., evolutionary information).

An approach, known as multiple kernel learning (MKL), has emerged to deal with different issues in machine learning, mainly, regarding support vector machines (SVM). The intuitive idea of MKL is that learning can be
enhanced when using different kernels instead of an unique kernel. Indeed, local analysis provided by each kernel is of benefit to examine the global structure of the whole data. From this idea, we introduce a dynamic kernel spectral clustering (DKSC) approach that is based on MKL. This approach uses the so-called kernel spectral clustering (KSC), which is based on KPCA formulation from least-square support vector machines and has shown to be a powerful tool for clustering hardly separable data allowing also out-of-samples extensions. MKL is used in such a manner that kernel matrices are computed from an input data sequence, in which each data matrix represents a frame at a different time instance. Afterwards, an accumulative kernel is calculated as a linear combination of the previously obtained kernels where the weighting factors are obtained by ranking each frame contained in the sequence. Such ranking is done by combining a relevance analysis procedure and a simple MKL approach.

Thu 25 - Thu 25 Apr-13 CIT Departemental lecture - WG1/WG3 seminar - Prof. Flavio Manenti
CIT 01.06
4:00 pm-5:00 pm
"Multi-scale modeling and optimization of sulfur recovery units"

Prof. Flavio Manenti,
Assistant Professor of Chemical and Industrial Processes and Plants at
the Dipartimento di Chimica, Materiali e Ingegneria Chimica “Giulio
Natta”, Politecnico di Milano

Abstract

Process simulation is nowadays supported by many tools and commercial flowsheeting packagesinvolving unit operations, reactors, thermodynamic libraries, and physical-chemical propertydatabases. These tools make possible the steady-state or dynamic simulation of complex processesand overall plants. Nevertheless, they still have key-open-issues to be handled to perform accurateand efficient simulations and to deepen the process understanding as well.One of the hardest problems is the simulation of non-ideal reactors via detailed kinetic schemes.This lack in current simulators is mainly due to three reasons: (1) the need of comprehensive andwell-established kinetic schemes to characterize the complex reaction environment; (2) the need toface simulation issues at different scales (kinetic, reactor, and plant scales); and (3) the need ofpowerful solvers to handle the resulting large-scale, stiff, strongly nonlinear systems that come fromthe detailed kinetic modeling.After a basic description of the process and the state of the art of its process modeling and industrialbest practice as well, the seminar deepens the detailed modeling of the different scales above. Thus,the main phenomena governing the behavior of the thermal section are discussed, together with thereaction mechanisms for the main species of the overall kinetic scheme (2400 reactions and 140radical/molecular species). Then, the kinetic scheme is integrated in a reactor network of idealreactors able to simulate more complex non-ideal reactor systems. Existing tools and numericalmethods are described and different industrial configurations are considered, comparing the modelpredictions with the industrial data. Finally, certain modeling and optimization problems arebroached at the process scale. Specifically, the formation of coking at the catalytic converters and thedata reconciliation and plant optimization under poor-redundancy measures are taken into account.

Flavio Manenti

He is assistant professor of chemical and industrial processes and plants at the Dipartimento diChimica, Materiali e Ingegneria Chimica “Giulio Natta”, Politecnico di Milano, where he receivedthe M.Sc. degree and the Ph.D. in chemical engineering. His research activities are on processmodeling and optimization, with 12 ongoing M.Sc. and Ph.D. student projects. He has co-authoredmore than 50 international publications on journals, proceedings, and book series and leaded morethan 20 R&D and technological transfer projects. He is the scientific coordinator of an internationalproject within the bilateral agreements with Argentina and the director of the REINFORCE(Renewable Energy and INtensification FOR Chemical procEsses) excellence cluster with 34partners. He is member of scientific committees of international congresses and associate editor ofan international peer-reviewed journal. He is responsible for undergraduate courses “Chemicalprocess dynamics and control”, “Chemical Plants” at Politecnico di Milano and lecturer atUniversità dell’Insubria.

Fri 19 - Fri 19 Apr-13 PhD defense - Tillmann Falck
Auditorium Oude Molen, Lokaal 00.07, Kasteelpark Arenberg 50, 3001-Heverlee
2:00 pm
Nonlinear System Identification using Structured Kernel Based Modeling

Tillmann Falck (KU Leuven, ESAT-SCD)

Abstract: This thesis discusses nonlinear system identification using kernel based models. Starting from a least squares support vector machine base model, additional structure is integrated to tailor the method for more classes of systems. While the basic formulation naturally only handles nonlinear autoregressive models with exogenous inputs, this text proposes several other model structures. One major goal of this work was to exploit convex formulations or to look for convex approximations in case a convex formulation is not feasible. Two key enabling techniques used extensively within this thesis are overparametrization and nonquadratic regularization. The former can be utilized to handle nonconvexity due to bilinear products. During this work overparametrization has been applied to handle new model structures. Furthermore it has been integrated with other techniques to handle large data sizes and a new approach to recover a parametrization in terms of the original variables has been derived. The latter technique, nonquadratic regularization, is also suitable to construct convex relaxations for nonconvex problems. In this context the major contribution of this thesis is the derivation of kernel based model representations for problems with nuclear norm as well as group-l1 norm regularization.

In terms of new or improved model structures, this thesis covers a number of contributions. The first considered model class are partially linear models which combine a parametric model with a nonparametric one. These models achieve a good predictive performance while being able to incorporate physical prior knowledge in terms of the parametric model part. A novel constraint significantly reduces the variability of the parametric model part. The second part of this thesis, that exploits structure to identify a more specific model class, is the estimation of Wiener-Hammerstein systems. The main contributions in this part are a thorough evaluation on the Wiener-Hammerstein benchmark dataset as well as several improvements and extensions to the existing kernel based identification approach for Hammerstein systems. Besides targeting more restricted model structures also several extensions of the basic model class are discussed. For systems with multiple outputs a kernel based model has been derived that is able to exploit information from all outputs. Due to the reliance on the nuclear norm, the computational complexity of this model is high which currently limits its application to small scale problems. Another extension of the model class is the consideration of time dependent systems. A method that is capable of determining the times at which a nonlinear system switches its dynamics is proposed. The main feature of this method is that it is purely based on input-output measurements. The final extension of the model class considers linear noise models in combination with a nonlinear model for the system. This work proposes a convex relaxations to estimate the noise model as well as a model capturing the system dynamics by solving a joint convex optimization problem. The final contribution of this thesis is a reformulation of the classical least squares support vector formulation that allows the analysis of existing models with respect to their sensitivity to perturbations on the inputs.

Promotor: Prof. Dr. Johan A.K. Suykens
Co-promotor: Prof. Dr. Bart De Moor

Tue 26 - Thu 28 Mar-13 32nd Benelux Meeting on Systems and Control
Houffalize, Belgium
9:00 am-5:00 pm
32nd Benelux Meeting on Systems and Control,
Houffalize, Belgium

March 26-28, 2013

http://saas.ulb.ac.be/bm2013/

Keynotes and courses by John Lygeros (ETH Zurich),
Jean-Christophe Poggiale (Aix-Marseille University)
Thierry Denoeux (Compiègne)
Franco Blanchini (Udine)

Wed 20 - Wed 20 Mar-13 WG3 Seminar - Michael Mangold
CIT 01.01
2:00 pm-3:00 pm
"Recent advances in model reduction for population balance systems"

Michael Mangold
Max Planck Institute for Dynamics of Complex Technical SystemsSandtorstrae 1, 39106 Magdeburg, Germany

Abstract

The evolution of particle populations, e.g. in crystallisation or polymerisation, is strongly a ffected by the flow conditions in the fluid phase (aggregation, breakage, attrition etc.). Process models that account for the relevant physical eff ects in an adequate way comprise Navier Stokes equations, energy balances, and mass balances for the liquid phase, as well as population balance equations for the particle phase. They are distributed in several external (space) and internal (property) coordinates and are computationally very demanding. A direct application of such models to process control and process design problems is infeasible due to the enormous computational burden. Hence there is a need for reduced control oriented models of low system order.
In this contribution, a model reduction technique for population balance systems coupled with fluid dynamics is presented [1, 2, 3]. A model of a laboratory scale urea crystalliser serves as an application example. A detailed reference model with two external and one internal coordinate implemented in OpenFOAM forms the starting point of the work. Proper Orthogonal Decomposition (POD) techniques are used to derive the reduced model. While POD is a well-established approach for model reduction of Navier Stokes equations, it has hardly been applied so far to particle systems. One of the additional difficulties compared to pure Navier Stokes equations are the more complicated nonlinear terms. An efficient and accurate treatment of the nonlinearities during runtime of the reduced model is non-trivial. It is found that the best point interpolation method recently introduced by Nguyen et al. [4] for parameterizedfunctions is an elegant way to solve this problem.
In an outlook, the application of the reduction technique to LDPE reactors is discussed. The fi nal objective is to model and control the formation of foulant layers.

References
[1] M. Krasnyk and M. Mangold. Reduction of a urea crystallizer model by proper orthogonal decom-position and best point interpolation. Industrial & Engineering Chemistry Research, 49:9887{9898,2010.
[2] M. Krasnyk, M. Mangold, and A. Kienle. Reduction procedure for parametrized uid dynamicsproblems based on proper orthogonal decomposition and calibration. Chemical Engineering Science,65:6238{6246, 2010.
[3] M. Krasnyk, M. Mangold, S. Ganesan, and L. Tobiska. Reduction of a crystallizer model withinternal and external coordinates by proper orthogonal decomposition. Chemical Engineering Science(accepted), 2011.
[4] N.C. Nguyen, A.T. Patera, J. Peraire. A \best-points" interpolation method for ecient approxima-tion of parametrized functions. Int. J. Numer. Methods Eng. 73:521{543, 2008.

Biographical information
Michael Mangold
1993 Diplom-Ingenieur degree in Technical Cybernetics from the University of Stuttgart
2000 Ph.D. degree from the University of Stuttgarttitle of thesis: "Nonlinear Analysis and Technical Application of Circulating Reaction Fronts"
since 1998 member of Max Planck Institute Magdeburg2006 Habilitation at Otto von Guericke University Magdeburgtitle of thesis: "Computer Aided Modeling, Analysis and Control of MembraneReactors and Fuel Cells"
2011 adjunct professor (auerplanmaiger Professor) for System Dynamics at theFaculty for Electrical Engineering and Information Technology of Otto vonGuericke University Magdeburg

Wed 20 - Wed 20 Mar-13 KU Leuven seminars on optimization in engineering - Andreas Argyriou
ESAT 00.57
2:00 pm
"Matrix Learning Problems and First-Order Optimization"

Andreas Argyriou (Ecole Centrale Paris)

In the past few years, there has been significant interest in nonsmooth convex optimization problems involving matrices, especially in the areas of machine learning, statistics and control. Instances of such problems are multitask learning and matrix completion, robust PCA, sparse inverse covariance selection etc. I will present PRISMA, a new optimization algorithm for minimizing a convex objective which decomposes into three parts: a smooth part, a simple non-smooth Lipschitz part, and a simple nonsmooth non-Lipschitz part. Our algorithm combines the methodologies of smoothing and accelerated proximal methods. Moreover, our convergence result removes the assumption of bounded domain required by Nesterov’s smoothing methods. I will show how PRISMA can be applied to the problems of max-norm regularized matrix completion and clustering, robust PCA and sparse inverse covariance selection, and compare to state of the art methods. Joint work with N. Srebro (TTI Chicago) and F. Orabona (TTI Chicago).

(OPTEC WG2 seminar)

Tue 19 - Tue 19 Mar-13 WG3 Seminar - Michael Mangold
CIT 01.01
3:00 pm-4:00 pm
"Recent advances in parameter identification and optimal experimental design"

Michael Mangold
Max Planck Institute for Dynamics of Complex Technical SystemsSandtorstrae 1, 39106 Magdeburg, Germany

Abstract

Virtually all mathematical models of chemical or biochemical processes contain unknown parameters that have to be identifi ed from experimental data. Parameter identi fication is therefore a central step during the development of mathematical models and a prerequisite for model based process control and process design.
Optimal experimental design (OED) in general tries to set experimental conditions that minimize uncer-tainties in the identi ed model parameter values, in the model structure, or in the model prediction. Onekey problem of OED is how to quantify these uncertainties. The most widely used approach employsthe Fisher information matrix. Although very efficient, this approach is based on a system linearisationand may give inaccurate results for nonlinear systems. The alternative to sample complete probabilitydistributions by Monte Carlo simulations is prohibitively expensive in most cases. Sigma point methodsmay be considered as a compromise between these two extremes, o ering higher accuracy than the Fisher information matrix while requiring less computation time than Monte Carlo methods.

The fi rst part of this presentation gives an overview on the application of sigma points to the solution of various problems in the area of OED. First, OED for parameter identi fication of a given model is considered [1]. It is shown that sigma points accurately predict the con fidence intervals of unknown parameters. Further the method can be used to compute global parameter sensitivities in connection with Sobol indices [2], or to compute and minimize confi dence intervals of the model states resulting from uncertainties in the model parameters [3]. Finally, application of sigma points to model discrimination problems [4] is discussed.
The second part of the presentation introduces a new method for parameter identification that is basedon inverse models and the system property of diff erential flatness. The treatment of noisy measurementsand the extension of the method to delay di fferential equations are discussed.

References

[1] R. Schenkendorf, A. Kremling, M. Mangold. Optimal experimental design with the sigma point method. IET Systems Biology, 3:10{23, 2009.
[2] R. Schenkendorf, M. Mangold. qualitative and quantitative optimal experimental design for parameter identi cation of a MAP kinase model In: Preprints of the 18th IFAC World Congress, pp. 11666{11671, 2011.
[3] R. Schenkendorf, A. Kremling, M. Mangold. Influence of non-linearity to the optimal experimental design demonstrated by a biological system. Mathematical and Computer Modelling of Dynamical Systems (accepted), 2012.
[4] R. Schenkendorf, M. Mangold. Optimal Experimental Design and Model Selection by a Sigma Point Approach.In: Proceedings MATHMOD2009 Vienna, 2009
[5] R. Schenkendorf, M. Mangold, U. Reichl. Parameter Identi cation Of Time-Delay Systems: A Flatness BasedApproach. In: Proceedings MATHMOD 2012 Vienna, 2012.

Biographical information
Michael Mangold
1993 Diplom-Ingenieur degree in Technical Cybernetics from the University of Stuttgart
2000 Ph.D. degree from the University of Stuttgarttitle of thesis: "Nonlinear Analysis and Technical Application of Circulating Reaction Fronts"
since 1998 member of Max Planck Institute Magdeburg2006 Habilitation at Otto von Guericke University Magdeburgtitle of thesis: "Computer Aided Modeling, Analysis and Control of MembraneReactors and Fuel Cells"
2011 adjunct professor (auerplanmaiger Professor) for System Dynamics at theFaculty for Electrical Engineering and Information Technology of Otto vonGuericke University Magdeburg

Tue 12 - Tue 12 Mar-13 SISTA Seminar - Marko Seslija
ESAT 00.62
2:00 pm
"Discrete Geometry Approach to Structure-Preserving Discretization of
Port-Hamiltonian Systems"

Marko Seslija (University of Groningen)

Abstract:
Computers have emerged as essential tools in the modern scientific
analysis and simulation-based design of complex physical systems. The
deeply-seated abstraction of continuity immanent to many physical
systems inherently clashes with a digital computer's ability of storing
and manipulating only finite sets of numbers. While there has been a
number of computational techniques that proposed discretizations of
differential equations, the geometric structures they model are often
lost in the process. In this talk, I will present a geometric framework
for the discretization of a class of physical systems (so-called
port-Hamiltonian systems) without destroying the underlying geometric
structure of the original system. The most important consequences of
such an approach are that many of the important results from
differential geometry can be transferred into the discrete realm and
thereby lead to numerically and physically faithful models.

Thu 7 - Fri 8 Mar-13 19th Belgian Mathematical Programming Workshop
Floréal Club, avenue de Villez, 6 in La-Roche-en-Ardennes
9:00 am-5:00 pm
"19th Belgian Mathematical Programming Workshop"
March 7-8, 2013.
Floréal Club, avenue de Villez, 6 in La-Roche-en-Ardennes.

Invited speakers:
Stan Van Hoesel (Maastricht University),
Marco Lübbecke (RWTH Aachen University)

More details can be found on http://sites.uclouvain.be/BMPW/

Mon 4 - Mon 4 Mar-13 "Nonsmooth optimization in machine learning"
University of Liege
10:30 am-4:30 pm

"Nonsmooth optimization in machine learning"

The workshop will take place in the academic room in the heart of the city. It will feature outstanding international speakers: Michael Jordan (UC Berkeley), Johan Suykens (KULeuven), and Francis Bach (ENS Paris/INRIA). The light format will allow for ample time of discussion and exchange throughout the day.

The workshop is sponsored by the FRFC project 'Nonsmooth manifold optimization" and by the IAP network "Dynamical Systems, Control and Optimization" DYSCO.

Registration is free but mandatory through the workshop website.

Wed 27 - Wed 27 Feb-13 KU Leuven Seminars on Optimization in Engineering - Maarten Breckpot
ESAT 00.57
3:00 pm-4:00 pm
"Flood control with Model Predictive Control for river systems with water reservoirs"

Maarten Breckpot
KU Leuven, Electrical Engineering Department (ESAT-SCD)

Abstract:
Many control strategies can be found in literature for controlling river systems. One of these methods is Model Predictive Control (MPC) and it has already shown its efficiency for set-point control of reaches and irrigation channels. In this presentation we will show that MPC can also be used for flood control of river systems. The proposed controller uses the buffer capacity of water reservoirs in an optimal way when there is a risk of flooding and it recovers the used buffer capacity as fast as possible. The performance of the controller is tested on a river system consisting of multiple channels, gates and a water reservoir. The controller will be used in combination with a Kalman filter which estimates all the states of the river system based on a very limited number of measured water levels. It was observed that the influence of this estimator on the control performance was minimal.

Tue 26 - Tue 26 Feb-13 KU Leuven Seminars on Optimization in Engineering - Volkan Cevher
ESAT 00.62
3:00 pm-3:45 pm
"Scalable quantum tomography via sparse projections onto the simplex"

Volkan Cevher,
Laboratory for Information and Inference Systems
EPFL Lausanne

Abstract

In this talk, we describe a scalable and accurate quantum tomography (QT) recovery framework with approximation guarantees. Our objective is to accurately recover a d x d complex positive semi-definite (PSD) matrix X with a known rank r from dimensionality reducing measurements as y=A(X)+e, where A is a subsampled Pauli operator, and e is some bounded perturbation.

The QT problem has three salient aspects that set it apart from the existing low-rank recovery problems:

1. Standard nuclear norm minimization approaches are not directly applicable in QT since X is a PSD density matrix and must have a trace of 1.

2. The Pauli measurement operator A in QT creates a major scalability bottleneck as the range of its adjoint is fully dense and the size of X is exponential in the number of quantum bits q (qubits): d=2^q.

3. The Pauli measurement operator A has the rank restricted isometry property (RIP) from O(rdlog^6d) measurements.

We show how to achieve provable QT recovery in linear space and quadratic time (in d) and computationally demonstrate 16-qubit recovery from 4rd Pauli measurements. To this end, we derive new sparse simplex projections, leverage randomized SVD’s, and propose a new online subspace reweighting technique. We also describe how our algorithmic developments apply to seemingly different problems, where convex relaxations of the sparsity and rank appear as constraints, such as measure learning, Markowitz portfolio optimization, and metric learning problems.

Wed 20 - Wed 20 Feb-13 KU Leuven Seminars on Optimization in Engineering - Mathieu Claeys
ESAT 00.62
3:00 pm-4:00 pm
"An introduction to the moment approach for optimal control"
Mathieu Claeys

This tutorial talk is devoted to the moment/sum-of-squares approach, which provides a general framework for global optimization of many problems, including optimal control.

In the first part of the talk, we will review the philosophy of the approach on the problem of global minimization of a polynomial function, for which the user-friendly GloptiPoly toolbox is readily available. The focus will therefore be on introducing the mathematical tools and the elegant transformations that are used behind the scene by the software: transformation of the original problem into a linear problem on measures, then manipulation of those measures by their moments to obtain tractable semi-definite relaxations.

The second part of the talk will show how to extend the approach for functional optimization problems such as optimal control. Measures arise quite naturally in this context, and the moment approach is therefore well suited to tackle those problems fruitfully. This will be shown on the two examples of bounded and impulsive control problems.

(slides)

Wed 20 - Wed 20 Feb-13 KU Leuven Seminars on Optimization in Engineering - Pierre Duysinx
MTM 03.19 (Kasteelpark Arenberg 44, B-3001 Heverlee)
2:00 pm-3:00 pm

"Shape and topology optimization of structural components: formulations and efficient solution algorithms"

prof. Pierre Duysinx (Department of Aerospace and Mechanical Engineering, ULg)

ABSTRACT

To face the challenge of higher performance and better energy efficiency, engineers call for efficient and rational tools to handle large and complex design problems. Combining computer aided design and mathematical programming algorithms provides an efficient and rationale approach to engineering design. Structural optimization has been developed now for more than 30 years with a growing success. Leading edge research is still pushing ahead the frontiers trying to solve larger scale problems, more general formulations, including more physics.

Initiated by the pioneer work of Bendsøe and Kikuchi (1988), topology optimization is one of the most flexible and general approaches. The design problem is formulated as an optimum material distribution. Because of the bitmap description of the geometry, the formulation introduces a large number of design variables, so that a global performance function is generally preferred in the design problem statement. An interesting contribution consists in extending the scope of the minimum compliance problem to local design restrictions such as local material failure. We are also investigating strongly non-linear problems arising in MEMS design problems dominated by coupled electro mechanical behaviors.

More recently, the (topology) optimal material description has been challenged by the novel level set description of the geometry. This implicit description of boundaries enables a flexible description of the geometry allowing a certain modification of the topology but with a smoother description of the shape. The level set description is naturally combined with the eXtended Finite Element Method (XFEM) to be able to relax the constraint of using conforming meshes, simplifying strongly the user interaction with the optimization process. The approach naturally enables a higher capability in handling various design constraints. For instance, level set description is applied to the shape optimization of flexible bodies subject to dynamics loading in complex multibody systems. Thus shape optimization using level set description provides a very nice complementary tool to topology optimization.

Finally a major characteristic of the research approach that is developed lies in the great attention paid to tailor efficient solution algorithms coping with the larger and larger scale optimization problems. The approach relies on the so-called sequential convex programming approach: the implicit and non-linear optimization problem is replaced by a sequence of explicit approximations. Then the convex sub-problems are solved by resorting to efficient math programming algorithms, in particular the dual maximization optimizer.

(slides)

Wed 13 - Wed 13 Feb-13 KU Leuven Seminars on Optimization in Engineering - Herman Bruyninckx
ESAT 00.62
2:00 pm-3:00 pm
"Best and worst practices in component-based software for complex engineering systems"

Herman Bruyninckx
Department of Mechanical Engineering,PMA, KU Leuven.

Everywhere in the world, research is aiming at making robots more interactive (with humans but also with other robots or cognitive agents'') and, at the same time, equip them with more and more powerful mobile and redundant manipulation and process capabilities. To control such systems, ever more complex and efficient constraint optimization solvers are required. But none of the existing software infrastructures (in automation and industrial robotics, but also in academic research) is yet flexible and complete enough to support the wide variety of requirements: functional, running on all kinds of embedded and accelerated hardware, configurability, multi-vendor, interconnected to knowledge bases, etc. This presentation gives an overview of some ongoing work in this direction, starting from an annotated summary of best and worst practices that appear in the robotics software community over the last decade.

Wed 6 - Wed 6 Feb-13 SISTA Seminar - Mustak Yalcin
ESAT 01.60
2:00 pm

Path Planning on Cellular Nonlinear Network using Active Wave Computing

Mustak E. Yalcin (Istanbul Technical University)

Active wave computing based algorithm for real-time robot navigation problem in dynamically changing environment has been already presented and tested in our previous works. A two- dimensional Cellular Neural/Nonlinear Network (CNN), consist of relaxation oscillators, has been used to generate the active wave in these works. In order to perform active wave computing based new algorithms, a practical network model (wave computer ) and its implementation are demanded. Therefore, in the first part of this talk, the wave computer and its implementation will be presented. In the second part, the algorithm to solve robot path finding problem using active wave computing techniques will be introduced. In order to improve the functionality of the active wave computing, novel techniques is required such as synchronization. In the third part of the talk, the Doppler Effect which provides a new qualification to the CNN-based wave computing techniques by putting the wave source’s motion into use will be introduced.

Tue 5 - Tue 5 Feb-13 Symposium 2013 Honorary doctorates
Auditorium Arenberg Castle (morning) + Auditorium Thermotechnisch Instituut (afternoon)
10:00 am-6:00 pm
Dear Colleague

We have the pleasure to invite you to the Symposium 2013 Honorary
doctorates 'brainstormers'  at ESAT- SCD-SISTA in honour of prof. Roska
and prof. Chua who are receiving a honorary doctorate on the occasion of
Patrons Saint day on February 4th at KU Leuven.

The symposium  will take place on Tuesday February 5th 2013.
http://homes.esat.kuleuven.be/~sistawww/eredoctoraat-brainstormers/index.php

Prof. Roska lecture entitled: Cellular Wave Computing for (artificial
and natural ) visual processing will take place in Auditorium Arenberg
Castle .

Prof. Chua lecture entitled: Memristor Hodgkin-Huxley , and Edge of
Chaos will take place in Auditorium Thermotechnisch Instituut .

A reception for all participants will follow in the Machinezaal
Thermotechnisch Instituut

If you would like to attend the lectures, please register on:

http://homes.esat.kuleuven.be/~sistawww/eredoctoraat-brainstormers/registration.php

We hope to welcome you on this unique special event.

Sincerely,

prof. dr. ir. Joos Vandewalle
Dept. Elektrotechniek, ESAT/SCD (SISTA)

Wed 30 - Wed 30 Jan-13 OPTEC Compact Course on Direct Multiple Shooting with YALMIP
ESAT 02.53
8:00 am-2:30 pm
OPTEC Compact Course on Direct Multiple Shooting with YALMIP

by Moritz Diehl, Rien Quirynen, Mario Zanon

Aim of this 6 hour course is to introduce PhD students with a
solid nonlinear and convex
optimization background to the field of continuous time
optimal control. The focus is on the direct Multiple Shooting (MS) method
of Bock and Plitt 1984 [Bock1984] in conjunction with
Sequential Convex Programming (SCP) [TranDinh2012b].

Main topic is the implementation of a
direct multiple shooting method in a MATLAB environment based
on the convex optimization solvers in YALMIP and fast integrators
from the ACADO Integrators Suite. Two application problems
are considered, a classical two state test example [Chen1998]
and a four state kite [Diehl2004f].

Course Schedule:

8:00 Lecture 1 by Moritz Diehl:
Convex Optimal Control and Sequential Convex Programming

8:30 Exercise 1:
Convex Optimal Control and Sequential Convex Programming

9:15 Lecture 2(a) by Mario Zanon:
The Classical Direct Multiple Shooting Method

9:45 Coffee Break in room 02.58

10:30 Lecture 2(b) by Rien Quirynen:
The ACADO Integrator Suite for Code Generated Integrators

11:00 Exercise 2:
Auto Generated Integrators, Direct Multiple Shooting, and SCP

12:00 Lunch break

13:00 Exercise 3:
Self Chosen Problems such as Non-Standard State and Parameter Estimation Problems, L1, Huber Penalty, Experimental Design, ...

14:00 Summary and Discussion of Preliminary Results

14:30 End

******

Solutions exercises:
Some MATLAB code for exercises 1 and 2 can be found here.

Course Registration:
Please fill in the following doodle for your registration and this as soon as possible.
http://www.doodle.com/tmik3arugg85abdu

Thu 24 - Thu 24 Jan-13 SISTA Seminar - Samuel Xavier-de-Souza
ESAT 00.62
3:00 pm
"Scalable Global Optimization with Coupled Simulated Annealing"

Prof. Samuel Xavier de Souza (Universidade Federal do Rio Grande do Norte)

Thu 17 - Thu 17 Jan-13 Simon Stevin Lecture on Optimization in Engineering - Chih-Jen Lin
Thermotechnical Institute, Auditorium van het 2de hoofdwet
4:00 pm-5:30 pm
25th Simon Stevin Lecture on Optimization in Engineering

"Optimization and machine learning "

Chih-Jen Lin
Department of Computer Science and Information Engineering,National Taiwan University

poster, flyer

Abstract

Optimization plays an important role in many machine learning methods. However, the two areas have very different focuses. The gap has caused that, on the one hand, some machine learning tasks may not suitably use optimization techniques, and on the other hand, optimization researchers may wrongly consider irrelevant issues when applying their methodology to machine learning. In this talk I will discuss my experiences on kernel and linear classification. In particular, we discuss support vector machines (SVM), which involve some challenging  optimization problems. As a machine learning researcher with an optimization background, I will show lessons learned in the past and how we eventually construct some widely used machine learning software.

Biographical Information

Chih-Jen Lin is a distinguished professor at the department of computer science and information engineering at the National Taiwan University. He studied mathematics at the National Taiwan University where he obtained his Bachelor degree in 1993. He received his Master degree and his PhD from the department of industrial and operations engineering  of the University of Michigan in 1996 and 1998 respectively.  After some time as a research associate in Argonne national laboratory, in 1998 he became a professor at the  National Taiwan University, where he is now a distinguished professor. Chih-Jen Lin has received numerous other awards for his research and is, among other, a fellow of the IEEE.

Prof. Chih-Jen Lin will also give 2 tutorials on January 15th and 16th. see

"Support vector machines and kernel methods: status and challenges" (link)

"Large-scale Machine Learning in Distributed Environments" (link)

The "Simon Stevin Lecture Series on Optimization in Engineering" is set up in order to promote optimization in engineering. For this aim, every quarter of the year an outstanding international scholar is invited to report on latest progress in the development of optimization algorithms and their applications in engineering.
Simon Stevin (1548-1620) was a Flemish mathematician and engineer. Among other, he helped to advance the use of decimal fractions, was the first to explain the tides by the attraction of the moon, and discovered the hydrostatic paradox. He made numerous inventions, among them a wind propelled carriage with sails, the "land yacht", which once impressed Prince Maurice of Orange as it moved faster than horses, in around 1600 on the beach between Scheveningen and Petten. Simon Stevin was fond of promoting the use of science in daily life and in craftmanship, and translated various mathematical terms into dutch. Among other, he introduced the dutch word for mathematics, "wiskunde".

Wed 16 - Wed 16 Jan-13 KU Leuven Seminars on Optimization in Engineering - Chih-Jen Lin
ESAT 00.62
9:30 am-12:00 pm
"Large-scale Machine Learning in Distributed Environments"

Chih-Jen Lin
Department of Computer Science and Information Engineering,National Taiwan University

Large-scale machine learning in distributed environments has emerged as an important research topic because data larger than a machine's capacity have become very common. We survey recent developments in this tutorial. First, we discuss when to and when not to apply distributed learning methods. Traditional machine learning algorithms focus on the computation, but we argue that in a distributed environment many other issues such as data locality must be taken into consideration. Second, we discuss distributed classification algorithms including linear and kernel Support Vector Machines (SVM), trees, and others. Third, we present approaches for distributed data clustering. Methods such as k-means, spectral clustering, and Latent Dirichlet Allocation (LDA) will be covered. Through the discussion of classification and clustering methods, we also see the advantages/disadvantages of different distributed programming frameworks such as MapReduce and MPI. Finally, we briefly discuss future challenges to tackle large-scale data classification and clustering.

Tue 15 - Tue 15 Jan-13 KU Leuven Seminars on Optimization in Engineering - Chih-Jen Lin
ESAT 00.62
9:30 am-12:00 pm
"Support vector machines and kernel methods: status and challenges"

Chih-Jen Lin
Department of Computer Science and Information Engineering,National Taiwan University

Support vector machines (SVM) and kernel methods are now important machine learning techniques. In this tutorial, we first introduce some basic concepts such as maximal margin, kernel mappings, and primal dual relationships. We then discuss the training by solving optimization problems and the practical use of SVM. Finally, we briefly mention some new research issues.

Thu 10 - Thu 10 Jan-13 SISTA Seminar - Kristiaan Pelckmans
ESAT 00.62
11:00 am
"Trading Recovery for Interpretability by Perturbing the Covariates"
Kristiaan Pelckmans (Uppsala University)

This work explores the effect of a heuristic to play up sparsity even further in traditional sparsity-promoting methods. The aim of this work is to provide interpretable solutions, even in cases where the experimental data does not satisfy the strong theoretical conditions (as RIP, null-space condition, etc.) which would guarantee recovery. A simple experiment is used to revisit some fundamental notions in the modern theory on compressed sensing. This work paves the way further to address problems of identifiability and robustness of modern data-driven methods for inferring interpretable models.

Wed 19 - Wed 19 Dec-12 KU Leuven Seminars on Optimization in Engineering - Roberto d'Ippolito
ESAT 91.91
10:30 am-11:30 am
"Large scale optimization for engineering systems - challenges and approaches"

Roberto d'Ippolito,
R&D Manager, Noesis Solutions, Belgium

Abstract:

The inherent complexity of MDO problems raises the problem of designing and optimizing high dimensional and multidisciplinary engineering systems: this problem is impacting design times and reliability of the results. Modern commercial CAE processes deployed to manage the design and development of products - whether these are automobiles, aircrafts or other products - are highly complex and use heterogeneous information coming from different disciplines. The multidisciplinarity and complexity of these processes is constantly increasing and numerical tools used in CAE to predict and improve products managed by these processes are exponentially growing. Although there are solutions being researched to face this huge computational challenge, the “brute force” computational approach will not solve all the problems. Powerful hardware, cloud computing and virtualization are not enough to address the challenges posed by complexity and heterogeneity alone: they need to be complemented by analysis techniques that are adequate to the new simulation demands. It is in this context that the potential of recent surrogate modelling techniques will play a major role. These techniques leverage dynamic sampling and model selection algorithms to efficiently exploit the knowledge extracted by direct simulation tasks and reuse this knowledge to build accurate and fast meta-models of system response. Recent advancements in this research area show in particular a potential to derive global models that are much faster to run, while keeping an equal (or even better) accuracy in the prediction of the system or component response.An overview of the challenges and recent advances made in this are will be given, together with indication of the most promising ones.

Wed 12 - Wed 12 Dec-12 KU Leuven Seminars on Optimization in Engineering - Warwick Graco
CIT 01.06
2:00 pm-3:00 pm

"ANALYTICS AND DATA MINING AT THE ATO"

Dr. Warwick Graco (Australian Taxation Office)

In the talk, Dr. Graco will discuss a few selected topics on data mining, risk analysis, social network analysis,... from a financial point of view.

Thu 15 - Thu 15 Nov-12 Simon Stevin Lecture on Optimization in Engineering - Volker Schulz
Thermotechnical Institute, Auditorium van het 2de hoofdwet
4:00 pm-5:30 pm

24th Simon Stevin Lecture on Optimization in Engineering

"Shape Optimization in PDE Based Applications"

Volker Schulz,
Department of Mathematics,University of Trier,Germany

slides

Abstract

Shape optimization is a challenging task, in particular in the context of PDE based applications. The talk will discuss recent approaches towards efficient optimization methods for applied shape optimization problems featuring a) a simultaneous optimization approach for the coupling of the treatment of the PDE with the optimization strategy and b) shape gradient and Hessian generation in the context of the shape calculus. Applications in the fields of aerodynamic, thermoelastic and acoustic applications will be presented. Furthermore, recent ideas towards an interpretation of the shape calculus in terms of optimization on the shape manifold will be discussed. These ideas have the potential to settle the annoying issue of the lack of symmetry of the standard shape Hessian and to enable a KKT point of view on shape optimization which has not been realizable so far.

Biographical Information

Volker Schulz is a full professor at the department of mathematics at the University of Trier, Germany.
He studied mathematics at the University of Augsburg where he obtained his diploma in 1990, and received his PhD and Habilitation from the University of Heidelberg in 1996 and 2000, respectively. After the PhD he was an associate researcher at the Institute for Computer Applications, University of Stuttgart, and in 1999, he became the head of the research group on Nonlinear Optimization and Inverse Problems at the Weierstrass Institute for Applied Analysis and Stochastics (WIAS) in Berlin. Since 2001 he is full professor in Trier, where he served, among other, head of the department, in 2002-2004. Volker Schulz' research interests are in several fields of numerical mathematics, in particular nonlinear large-scale optimization, partial differential equations, multigrid methods and inverse problems, with applications for example in robotics, chemical engineering, groundwater flow, and turbomachinery design. He is a member of the ediorial boards of the journals Computing and Visualization in Science, Acta Applicandae Matematicae, and the SIAM Journal on Scientific Computing, among other.

The "Simon Stevin Lecture Series on Optimization in Engineering" is set up in order to promote optimization in engineering. For this aim, every quarter of the year an outstanding international scholar is invited to report on latest progress in the development of optimization algorithms and their applications in engineering.
Simon Stevin (1548-1620) was a Flemish mathematician and engineer. Among other, he helped to advance the use of decimal fractions, was the first to explain the tides by the attraction of the moon, and discovered the hydrostatic paradox. He made numerous inventions, among them a wind propelled carriage with sails, the "land yacht", which once impressed Prince Maurice of Orange as it moved faster than horses, in around 1600 on the beach between Scheveningen and Petten. Simon Stevin was fond of promoting the use of science in daily life and in craftmanship, and translated various mathematical terms into dutch. Among other, he introduced the dutch word for mathematics, "wiskunde".

Participation is free but please register via the doodle:

http://www.doodle.com/d3fp3x84f98w83zt

Wed 14 - Wed 14 Nov-12 KU Leuven Seminars on Optimization in Engineering - Sergio Lucia
ESAT 00.57
11:00 am-12:00 pm
Non-conservative Robust Nonlinear Model Predictive Control by Multi-stage Optimization

Sergio Lucia
TU Dortmund (Chair for Process Dynamics and Operation)
Germany

Model Predictive Control (MPC) has become one of the most popular control techniques in the process industry mainly because of its ability to deal with multiple-input-multiple-output plants and with constraints. However, in the presence of model uncertainties and disturbances its performance can deteriorate. Therefore, the development of robust MPC techniques has been widely discussed during the last years, but rarely applied in practice due to the conservativeness or computational complexity of the approaches. In this talk, a robust non-conservative nonlinear model predictive control is presented. The approach is based on the representation of the evolution of the uncertainty as a scenario tree, leading to a non-conservative robust control of the uncertain plant. Simulation results show that multi-stage NMPC outperforms standard and min-max NMPC under the presence of uncertainties for a semi-batch polymerization benchmark problem. In addition, the advantages of the approach are illustrated for the case where only noisy
measurements are available and the unmeasured states and the uncertainties have to be estimated using an observer. Finally, the open problems and the main possibilities for future work are discussed.

Thu 8 - Fri 9 Nov-12 OPTEC Retreat 2012
Belgian coast
10:00 am-4:00 pm
OPTEC Retreat 2012

Internal Workshop where OPTEC Members meet
as a whole group and within their workgroups and
give talks to each other, to generate new ideas and
foster future cooperations. The location will probably be a place
at the Belgian coast.

The pdf printout of the OPTEC booklet can be find:
http://homes.esat.kuleuven.be/~optec/files/Booklet_final_corrected_list.pdf

Wed 31 - Wed 31 Oct-12 HIGHWIND Seminar - Alula Energy Concept and Airborne Wind Energy Activity in Finland
Arenberg Castle (room 00.0006)
1:30 pm-2:00 pm
Alula Energy Concept and Airborne Wind Energy Activity in Finland

Ilpo Suominen
Alula Energy, Finland

Abstract:
The Airborne Wind Energy research in Finland has traditionally been concentrated in the Mathematics Department of Tampere University of Technology. However, in summer 2011 Alula Energy, a spin off company from Tampere University of Technology, started to develop its unique concept of a landing and takeoff system for airborne wind energy power plants. Recently, academic airborne wind energy research has expanded to the University of Oulu. The airborne wind energy community in Finland and Alula Energy in particular believe strongly in collaborative R&D and supply chain thinking.

Tue 30 - Tue 30 Oct-12 Quarterly Meeting of OPTEC WG 4
C300.02.52 (Sem A)
2:00 pm-4:15 pm
"Quarterly Meeting of OPTEC Working Group 4 on PDE- constrained optimization"

Agenda:

14.00-14.30h: Presentation by Ingrid Lepot from Cenaero

14.30-14.45H: discussion and time for questions

14.45-15.00h: coffee break

Around 15h: end of the visit from Cenaero

15.15h: WG4 meeting continued (may start a bit earlier – depending on visit Cenaero)

15.15-15.45h: Overview of the two-slide presentations of the WG4 members (in preparation of the OPTEC retreat – cf. further below)

15.45-16.15h: Further planning of the OPTEC retreat – decision on agenda, possible presentations, …

Tue 30 - Tue 30 Oct-12 KU Leuven Seminars on Optimization in Engineering - Ingrid Lepot
C300.02.52 (Seminarie A)
2:00 pm-3:00 pm
"Multi-disciplinary surrogate-assisted optimization activities at Cenaero"

Ingrid Lepot
Cenaero, Walloon research center for Aeronautics, http://www.cenaero.be/

Abstract
In order to answer industrial design needs, Cenaero focuses on the one hand on the development of  online surrogate-based optimization algorithms and on the other hand on the development of tailored multi-disciplinary conception methodologies.  Both aspects will be covered in the presentation, with an overview of recent developments and research axis in terms of surrogates models and search processes as well as selected applications, ranging from the design of turbomachinery components and systems to biomedical engineering.

Wed 3 - Wed 3 Oct-12 Quarterly Meeting of OPTEC WG 1
ESAT 01.60
1:30 pm-4:00 pm
"Quarterly Meeting of OPTEC WG1 on Dynamic and Embedded Optimization"

Agenda:

13:30 Introduction of new members and visitors
13:45 Attila Kozma (ESAT): Comparing Distributed QP Solution Methods
14:10 Discussion
14:20 Sonja Rauski (Bremen): Large Scale Optimization with the SQP Method WORHP
14:40 Discussion
14:45 Break
15:15 Vryan Palma (Bayreuth): Sensitivity-based multistep feedback nonlinear MPC
15:40 Discussion
15:45 Wrap Up
Abstracts:

13:45
Comparing Distributed QP Solution Methods
Attila Kozma (KU Leuven)(slides)

We present   different methods to solve quadratic programs (QP) in a distributed manner
and compare their numerical performance and parallelizability
at hand of a benchmark collection. Joint work with C. Conte, M. Morari, M. Diehl.

***
14:20
Large Scale Optimization with the SQP Method WORHP
Sonja Rauski (Univ. Bremen and Astos Solutions)(slides)

Nonlinear optimization has grown to a key technology in many areas of the industry, especially for solving discretized optimal control problems with ODEs, DAEs and PDEs, which result in large and sparse problems. To solve these one has to exploit as much information of the problem as possible. This includes an efficient storing of the occurring matrices and vectors and an appropriate approximation of the Hessian. In this talk I will present some update techniques, introduced by Broyden-Fletcher-Goldfarb-Shanno (BFGS), used for approximation of the Hessian matrix, as well as some new ideas for improvement of these techniques. The solver WORHP will be introduced, which is a mathematical software library for solving continuous large scale nonlinear optimization problems numerically.

****

15:15
Sensitivity-based multistep feedback nonlinear MPC
Vryan Palma (Univ. Bayreuth) (slides)

A main challenge in NMPC applications is reducing the computational load
due to the online solution of an OCP at every time step. This task should
be accomplished without sacrificing statements on stability, performance
and robustness of the resulting control algorithm.

A straightforward approach is through multi-step feedback, i.e., using
more than just the first element of the resulting finite horizon optimal
control sequence and thus performing optimization less often. Upon using
longer control horizons, stability and performance results still remain
valid. However, this may reduce robustness since the use of more elements
of the control sequences suggests that the system runs in open loop for a
longer time wherein the multi-step feedback law is not able to account for
disturbances for which the states are susceptible to.

A remedy for this issue is incorporating sensitivity analysis to update
the next entry of the multi- step feedback, injecting the updated control
to the system to generate the next state, and repeating this process to
the succeeding entries of the multi-step feedback before finally
performing the next optimization solving the next nonlinear programming
problem.

Wed 3 - Wed 31 Oct-12 VUB Leerstoel 2012-2013 Data-driven modelling: an integrative approach- Prof. Johan Suykens
Promotiezaal D.2.01
3:00 pm-5:00 pm
VUB Leerstoel 2012-2013
Data-driven modelling: an integrative approach
Prof. Johan Suykens (KU Leuven, ESAT-SCD)

Program:
1. Advanced data-driven black-box modelling (1 hour, inaugural lecture, Oct 3: 16:00-17:00, Promotiezaal D.2.01)

2. Support vector machines and kernel methods in systems, modelling and control (2 hours, Oct 10: 15:00-17:00, Promotiezaal D.2.01)

3. Data-driven modelling for biomedicine and bioinformatics (2 hours, Oct 17: 15:00-17:00, Promotiezaal D.2.01)

4. Kernel methods for exploratory data analysis and community detection (2 hours, Oct 24: 15:00-17:00, Promotiezaal D.2.01)

5. Complex networks, synchronization and cooperative behaviour (2 hours, Oct 31: 15:00-17:00, Promotiezaal D.2.01)

The slides can be found at: http://www.esat.kuleuven.be/sista/members/suykens.html

Biography:

Johan A.K. Suykens is a Professor with KU Leuven. He is author of the books "Artificial Neural Networks for Modelling and Control of Non-linear Systems" (Kluwer Academic Publishers) and "Least Squares Support Vector Machines" (World Scientific) and co-author of the book "Cellular Neural Networks, Multi-Scroll Chaos and Synchronization" (World Scientific). He is a Senior IEEE member and has served as associate editor for the IEEE Transactions on Circuits and Systems (1997-1999 and 2004-2007) and for the IEEE Transactions on Neural Networks (1998-2009). He received an IEEE Signal Processing Society 1999 Best Paper (Senior) Award and several Best Paper Awards at International Conferences. He is a recipient of the International Neural Networks Society INNS 2000 Young Investigator Award for significant contributions in the field of neural networks. He has served as a Director and Organizer of the NATO Advanced Study Institute on Learning Theory and Practice (Leuven 2002), as a program co-chair for the International Joint Conference on Neural Networks 2004 and the International Symposium on Nonlinear Theory and its Applications 2005, as an organizer of the International Symposium on Synchronization in Complex Networks 2007 and a co-organizer of the NIPS 2010 workshop on Tensors, Kernels and Machine Learning. He has been recently awarded an ERC Advanced Grant 2011.

Wed 26 - Wed 26 Sep-12 A-DATADRIVE-B progress meeting
ESAT AUD B
10:00 am-4:00 pm
10:00 "Supervised Novelty Detection"
Vilen Jumutc

10:30 "L0-Norm Based Variations to LSSVM for Large Scale Data"
Raghvendra Mall

11:00 "Kernel spectral clustering with memory for studying evolving networks"
Rocco Langone

11:30 "Eigenvector re-usage in QP problems"
Gervasio Puertas

12:00 Lunch

14:00 "Gradient based methods for Lp regularization"
Andreas Argyriou

14:30 "Support Vector Machine with Pinball Loss"
Xiaolin Huang

15:00 "Regression with l1-regularization and Gaussian Kernels"
Lei Shi

15:30 "Prediction of QoS compliance of service compositions"
Dries Geebelen

--- Abstracts ---

- "L0-Norm Based Variations to LSSVM for Large Scale Data"

Least squares support vector machines (LSSVM) have found applications in the field of classification and regression resulting in performance comparable to support vector machines (SVM). Some of the drawbacks of LSSVM is the lack of
sparseness in the final model and inability to handle large scale data due computational costs (O(N^3) time complexity for solving a system of linear equations and O(N^2) memory requirement ). Recently, a Fixed-Size Least Squares Support
Vector Machine (FS-LSSVM) was proposed to introduce sparsity by using Nystr¨om approximations of the dataset using a set of prototype vectors (PV) and solving an over-determined system of linear equations. This set of PV is obtained by maximizing the quadratic R´enyi entropy of the dataset and results in a parametric model. However, a large number of PV are required to obtain comparable accuracy to the LSSVM. We introduce a second level of sparsity by performing an iterative approximation of L0-norm on the PV and identifying the essential set of prototype vectors namely SV required for building the sparse FS-LSSVM model. We propose another approach to introduce sparsity in LSSVM model by selecting the initial set of PV by means of quadratic Renyi entropy as representative points of the dataset. We perform training and validation on these PV using the LSSVM dual formulation, followed by selection of essential prototype vectors namely SV by an iterative sparsifying L0-norm. These selected appropriate SV are used for building a sparser LSSVM model resulting in a non-parametric model. The approximation in this approach is on set of points over which training and validation is performed to obtain the tuning parameters of the model. It overcomes the problem of memory constraints O(N^2) and computational costs O(N^3) as faced by current methods, resulting in a sparse LSSVM model. The underlying
approximations of the two models allow to scale the models to very large datasets. Experiments on real world classification and regression datasets from UCI repository illustrate that these adaptations achieve highly sparse models  without a significant trade off in error estimations and are applicable to large scale datasets.

- "Kernel spectral clustering with memory for studying evolving networks"

This work is related to the problem of community detection of evolving
networks, which for instance arises in the segmentation of moving
objects, clustering of telephone traffic data etc. A desirable feature
of a clustering model which has to capture the evolution of communities
over time is the temporal smoothness between clusters in successive
timesteps. In this way the model is able to track the long-term trend
and in the same time it smoothens out short-term variation due to noise.
We use the Kernel Spectral Clustering with Memory effect (MKSC) model,
which is based on a constrained optimization formulation typical of
Least Squares Support Vector Machines (LS-SVM). At the primal level the
objective function is designed to explicitly incorporate temporal
smoothness as a valid prior knowledge.

- "Gradient based methods for Lp regularization"

Motivated by learning problems, such as feature selection, multitask
learning and multiple kernel learning,
we propose a novel optimization algorithm for Lp regularization
problems. In the cases 1 < p < 2 and p > 2,
popular algorithms, such as forward-backward splitting or Nesterov's
methods, cannot be easily applied,
since proximity operators or projections cannot be computed with a
direct method. For this reason, we propose a very different
approach requiring only computations of gradients, which have a simple
closed form for Lp problems. We prove convergence of the iterates
and O(1/k) convergence rates for the objective, where k is the number
of iterations. This algorithm can also
be seen as an instance of a more general one, which applies to a large
class of optimization problems and is connected
to the Frank-Wolfe method for quadratic programming from the 1950s.

- "Support Vector Machine with Pinball Loss"

Traditionally, hinge loss, which is related to minimal value of a set, is used to construct support vector machine (SVM) but the result is sensitive to noise. By comparison, pinball loss is related to quantile and the result is less sensitive. Though the great statistical property of quantile regression makes pinball loss has been deeply studied and widely applied in regression field, pinball loss has not been used in classification. In this paper, we propose a SVM with pinball loss, of which the properties are investigated, including the misclassification error and the bounded influence function. Sensitivity curve and testing precision on numerical experiments both show the effectiveness of pinball loss SVM, which will be a promising tool for classifying noise corrupted data.

- "Regression with l1-regularization and Gaussian Kernels"

The quantile regression problem is considered by learning schemes based on
$\ell_1-$regularization and varying Gaussian kernels. Our analysis
shows that the convergence behavior of this algorithm is almost the same as
that of the RKHS-based learning schemes. Furthermore, the previous analysis for kernel-based quantile regression usually requires that the output sample values are uniformly bounded, which excludes the common case with Gaussian noise. Our error analysis can lead to satisfactory convergence rates even for unbounded sampling processes.

Thu 20 - Thu 20 Sep-12 KU Leuven Seminars on Optimization in Engineering - Anna Marconato
ESAT 02.58
2:00 pm

Title: “Combining system identification and statistical learning in nonlinear data-driven modeling”

Anna Marconato (VUB, ELEC)

Abstract: Most real-life systems are characterized by a nonlinear (dynamic) behavior. Many different techniques have been developed to solve system identification problems, and while the linear case is already well understood, nonlinear modeling seems still to offer interesting challenges and open problems. In this talk the problem of identifying nonlinear dynamic systems is addressed by combining the best of two worlds. On one hand, classic system identification offers a variety of widely accepted tools to model system dynamics. On the other hand, the statistical learning community provides us with powerful regression techniques to describe nonlinearities. A combination of these ideas is used in this work to develop a fast and efficient data-driven method for the initialization of nonlinear state-space models. More in details, the approach discussed here is based on the idea of transforming the general nonlinear dynamic problem into an approximate static formulation, by cutting the recursion in the state equation. Static regression methods can then be easily applied to separately estimate the nonlinear terms in the model. In this presentation, neural networks are used to illustrate the approach, and the possibility to include other choices of nonlinear functions (e.g. SVMs and LS-SVMs) is discussed. A number of benchmark problems (Silverbox, Wiener-Hammerstein, crystal detector, ...) are considered to validate the performance of the method on real measurement examples, and to perform a comparison with other existing nonlinear system identification techniques.

(OPTEC WG2 Data-driven modelling)

Wed 19 - Wed 19 Sep-12 SISTA Seminar - Ivan Markovsky
ESAT 02.58
11:00 am
"Hankel structured low-rank matrix completion"

Ivan Markovsky (University of Southampton)

Identification of a discrete-time linear time-invariant system from exact finite trajectory with missing values is equivalent to a Hankel structured low-rank matrix completion problem. There is no efficient solution to the latter problem, however, special cases (for example, all missing values located in the lower right corner of the matrix) have meaningful interpretations from a system theory point of view (partial realization, in the example) and some of them have efficient solution methods. The nuclear norm heuristic (replacing the rank constraint by the sum of the singular values) leads to a semidefinite programming problem that can be effective in solving small size structured matrix completion problems. An alternative approach motivated from ideas in subspace identification is also presented.

Tue 4 - Tue 4 Sep-12 SISTA Seminar - Dana Lahat
ESAT 00.62
11:30 am-12:30 pm

"Second-Order Multidimensional ICA: Theory and Methods"

Dana Lahat (Tel-Aviv University)

Abstract:

Independent component analysis (ICA) and blind source separation (BSS) deal with extracting a number of mutually independent elements from a set of observed linear mixtures. Motivated by various applications, this talk considers a more general and more flexible model: the sources can be partitioned into groups exhibiting dependence within a given group but independence between two different groups. We argue that this is tantamount to considering multidimensional components, as opposed to the standard ICA case which is restricted to one-dimensional components.

Multidimensional data may occur due to various complex relations within the dependent elements. The dimension of a dependent group may not always reflect the actual number of its underlying elements. As a result, in multidimensional models, there is not always a physically meaningful interpretation to separating the multidimensional components back into single-dimensional elements.

The core of this work is the statistical analysis of the blind separation of multidimensional components based on second-order statistics, in a piecewise-stationary model. We develop the likelihood and the associated estimating equations for the Gaussian case. We obtain closed-form expressions for the Fisher information matrix and the Cramér-Rao lower bound (CRLB) of the de-mixing parameters, as well as the mean square error (MSE) of the component estimates. For Gaussian data, our separation criterion achieves, up to higher-order terms, the CRLB, and is thus optimal in the MSE sense.

We present necessary and sufficient conditions for the model to be identifiable. These are also the sufficient and necessary conditions for joint block diagonalization (JBD) of any set of real positive-definite symmetric matrices to be unique.

We then turn to the case when the separation procedure is based on a one-dimensional model, followed by a clustering step, in which the one-dimensional output is assigned into groups, representing the multidimensional components. This attitude is common practice in various applications. We prove that for piecewise stationary data, and when only second-order statistics are used, this form of separation is suboptimal. In particular, we obtain a closed-form expression for the MSE of this separation procedure, an expression which is based only on the model parameters. By comparing this expression with the MSE when the correct model is used, one can obtain the exact expected gain directly from the model parameters, without resorting to numerical simulations or Monte-Carlo trials.

Our analysis is verified through numerical experiments. In addition, we demonstrate the theoretical gain in the accuracy of component recovery in the presence of multidimensional components for several dependence scenarios.

Fri 6 - Fri 6 Jul-12 WG3 seminar - Prof N. Abo-Ghander
CIT PW 00.01
11:00 am-12:00 pm
COUPLING DEHYDROGENATION OF ETHYLBENZENE WITH HYDROGENATION OF NITROBENZENE IN AN AUTOTHERMAL CATALYTIC MEMBRANE REACTOR

Prof N. Abo-GhanderKing Fahd University of Petroleum and Minerals, Saudi-Arabia

Dehydrogenation of ethylbenzene and hydrogenation of nitrobenzene form an interesting pair of reactions to be coupled in a catalytic membrane reactor. The former is reversible and  thermodynamically limited, supplying hydrogen with a net endothermality, while the latter is irreversible and exothermic, consuming hydrogen to produce aniline. In this work, coupling of these two reactions is simulated in a catalytic fixed bed membrane reactor where hydrogen produced on the dehydrogenation side is transferred through hydrogen membranes to the hydrogenation side where it reacts to produce aniline. Heat generated on the hydrogenation side is transferred to the dehydrogenation side, where it is utilized by the endothermic dehydrogenation reaction to improve the styrene yield.

A pseudo-homogeneous model for the coupled reactor based on the concept of fixed bed reactors accounting for both the diffusion of hydrogen and transfer of heat is first developed. The effects of the operating and design parameters considered on the production of styrene and aniline show conflicting behaviour, i.e. improving the yield of styrene results in decreased production of aniline. Consequently, the co-current configuration of the coupled reactor was optimized within constraints so that it can be operated effectively to produce ~98% styrene as a one limiting option or ~80% aniline at the other extreme.

The intraparticle diffusion resistance, a major limitation in fixed bed reactors, is evaluated by developing a heterogeneous reactor model based on Fickian diffusion and the dusty gas model for both isothermal and non-isothermal catalyst pellets. Both heterogeneous models predict a significant reduction in yield and conversion relative to the pseudo-homogeneous model, indicating the importance of heterogeneity. This reduction is generally less severe for the dusty gas model than for Fickian diffusion. The mean square deviation and absolute deviation along the reactor are calculated for all models relative to the heterogeneous reactor model with dusty gas for non-isothermal catalyst pellets, considered to be the most rigorous model tested. Assuming isothermality causes larger deviations than assuming Fickian diffusion. The deviations in the predictions of the homogenous model and the heterogeneous models from those of the dusty gas model for non-isothermal pellets are ~6% and ~11%, respectively.

Wed 4 - Wed 4 Jul-12 Simon Stevin Lecture on Optimization in Engineering - Annick Sartenaer
Auditorium of the Arenberg Castle
4:00 pm-5:30 pm
23rd Simon Stevin Lecture on Optimization in Engineering

"Data assimilation in oceanography and weather forecasting"

Annick Sartenaer
Facultés Universitaires Notre-Dame de la Paix (FUNDP), Namur (Belgium)

poster, flyer,

Abstract

Data assimilation is a methodology for estimating the initial state of a dynamical system by combining the information from observational data and from a numerical prediction model that describes the evolution of the system. The most important fields of application of data assimilation are the ocean and weather forecasts. In this talk, we briefly survey the two main approaches used in data assimilation: the sequential one, based on the statistical estimation theory (Kalman filter) and the variational one, based on the optimal control theory. This last approach amounts to solve a very large weighted nonlinear least-squares problem called 4D-Var (four-dimensional variational problem). Focusing on the solution of the 4D-Var problem, we discuss two challenging issues in the context of large-scale operational data assimilation: preconditioning techniques, for accelerating the convergence, and derivative-free techniques, to avoid the computation of derivatives.

Biographical Information

Annick Sartenaer is full professor at the Mathematics Department of the University of Namur (FUNDP) in Belgium, from which she received her PhD in 1991.  She was Senior Researcher at CERFACS (European Center for Research and Advanced Training in Scientific Computation) in France from 1997 to 1999 and Research Associate of the FNRS (Fond National de la Recherche Scientifique) from 1999 to 2002. She received her Habilitation (HDR) from the University Paul Sabatier in Toulouse in 1999.  Her research interests are in the area of nonlinear continuous optimization, with a special emphasis on large-scale issues, and in real-world applications such as weather forecast or energy distribution.

The "Simon Stevin Lecture Series on Optimization in Engineering" is set up in order to promote optimization in engineering. For this aim, every quarter of the year an outstanding international scholar is invited to report on latest progress in the development of optimization algorithms and their applications in engineering.
Simon Stevin (1548-1620) was a Flemish mathematician and engineer. Among other, he helped to advance the use of decimal fractions, was the first to explain the tides by the attraction of the moon, and discovered the hydrostatic paradox. He made numerous inventions, among them a wind propelled carriage with sails, the "land yacht", which once impressed Prince Maurice of Orange as it moved faster than horses, in around 1600 on the beach between Scheveningen and Petten. Simon Stevin was fond of promoting the use of science in daily life and in craftmanship, and translated various mathematical terms into dutch. Among other, he introduced the dutch word for mathematics, "wiskunde".

Tue 19 - Tue 19 Jun-12 KU Leuven Seminars on Optimization in Engineering - Francesco Orabona
ESAT 00.62
2:00 pm
(OPTEC - WG2 seminar)

"Efficient Stochastic and Batch Optimization Algorithms"
Francesco Orabona (TTI Chicago)

Many machine learning problems reduce to solving a convex
optimization problem. The choice of the correct optimization algorithm
is often critical to obtain a good solution in a reasonable amount of
time, and it depends on the characteristics of the objective function.
I present a stochastic and a batch algorithm for efficiently solving
strongly convex functions and composite ones with Lipschitz parts
respectively. I also show how the optimization algorithm can guide the
design of the objective function, to achieve good performance and fast
convergence rates. Experimental results will be shown in Multi Kernel
Learning, Matrix Completion and Robust PCA applications.

Thu 7 - Thu 7 Jun-12 Quarterly Meeting of OPTEC WG 4
MECH Baekeland room
2:00 pm-4:00 pm

"Quarterly Meeting of OPTEC Working Group 4 on PDE- constrained optimization"

Proposed agenda

1. Brainstorm on connection between applied math and application: try to define simple problems for CS (connect to exercises? Simple paper?) & Continue discussion on flow problems (Burgers equation etc.). (40 min)

2. Possible visit & presentation by Axel Kroener in Autumn (question from Moritz) (10 min)

3. next dates WG4 meeting + brief discussion OPTEC retreat (10 min)

4. coffee (10 min)

5. Brainstorm on next year’s aims and objectives for WG4 (30 min)

6. planning next meeting (10 min)

Thu 31 - Thu 31 May-12 Quarterly Meeting of OPTEC WG 3
BOKU 3.12
9:00 am-11:00 am
"Quarterly Meeting of OPTEC Working Group 3 on Parameter and State Estimation"

The preliminary agenda for this meeting is the following:
- Presentation by Kristof Maes
- Presentation by Joost Lauwers

Mon 21 - Mon 21 May-12 KU Leuven Seminars on Optimization in Engineering - Jens Lang
CS 200A 05.001
4:00 pm-5:00 pm
"Adaptive multilevel methods for large-scale optimal control problems"

Prof. Jens Lang
Technische Universität Darmstadt, Department of Mathematics
Numerical Analysis and Scientific Computing

In this talk, I will summarize our recent activities in solving time-dependent PDE-constrained optimal control problems with possible restrictions on control and state. I will first start with an introduction including general settings, first order optimality conditions, first-optimize-then-discretize as well as first-discretize-then-optimize approaches. This will be followed by a discussion of how to use a well designed sequence of adaptive meshes in space and time to efficiently approximate the infinite dimensional optimal control solution. I will also shortly discuss the construction of discrete adjoint time integrators. A three-dimensional glass cooling problem governed by radiative heat transfer is used to demonstrate the performance of the multilevel methods.

Fri 4 - Fri 4 May-12 SISTA Seminar - Kristiaan Pelckmans
ESAT 01.60
11:00 am
"Aggregated Prognosis Through Exponential Reweighting: A Case Study in the analysis of Micro-array data"

Kristiaan Pelckmans (Uppsala University)

ABSTRACT: This presentation discusses an application of machine learning and aggregation strategies to risk and survival analysis for high-dimensional problems. Theoretical evidence is found in studies of statistical and learning theory on aggregation strategies, and in online learning with expert advice. Such approaches do not only come with exciting theoretical merits, but also lead us to computationally efficient procedures. This in turn opens opportunities for dealing with extremely high-dimensional data produced e.g. by studies of Genome-Wide Association (GWA). Empirical evidence that such approach may outperform techniques based on empirical risk minimization and on common techniques borrowed from the traditional analysis of such data, are found in studies of micro-arrays for relating genetic signatures to risk analysis and prognosis of breast cancer.

Wed 2 - Thu 3 May-12 OPTEC Seminar on Tensors, Computing, Optimization and Signal Processing
CS 200A02.112 and ESAT 00.62
2:30 pm
OPTEC Seminar on Tensors, Computing, Optimization and Signal Processing

May 2-3, KU Leuven

Program (see below for detailed abstracts)

May 2

Dept. of Computer Science, 200A02.112

14h30-14h50: Lieven De Lathauwer, A Short Introduction to Tensor Decompositions

14h50-15h45: Eugene Tyrtyshnikov, Kronecker Representation of TT Decompositions, New Wavelet Transforms, Fast Tensor Interpolation

15h45-16h00: break

16h00-16h30: Laurent Sorber, State of the Art Overview of Algorithms for Tensor Decompositions

16h30-17h00: Nick Vannieuwenhoven, A New Truncation Strategy for the Higher-order Singular Value Decomposition

May 3

Dept. of Electrical Engineering (ESAT), 00.62

14h30-15h30: Boris N. Khoromskij, Efficient Numerical Approximation of Multi-dimensional PDEs in Quantized Tensor Spaces

15h30-15h45: Stefan Vandewalle, Overview Research NATW

15h45-16h00: break

16h00-16h30: Ignat Domanov, On the Uniqueness of the Canonical Polyadic Decomposition: New Sufficient Conditions Based on Properties of Khatri-Rao Products of Compound Matrices

16h30-17h00: Mikael Sorensen, Canonical Polyadic Decomposition with Orthogonality Constraints

17h00-17h30: Toon Van Waterschoot, Sensing Wave Fields in a Finite Element Framework

Lieven De Lathauwer, A Short Introduction to Tensor Decompositions

(slides)

We give some basic definitions from multilinear algebra and briefly discuss some fundamental tensor decompositions. We give a quick preview of their interest for computing, optimization, signal processing and large-scale data analysis.

Eugene Tyrtyshnikov, Kronecker Representation of TT Decompositions, New Wavelet Transforms, Fast Tensor Interpolation

Classical tensor decompositions (CP, Tucker) are important as data modelsbut not good enough as a base for fast tensor algebra algorithms. Most relevant tensor decompositions for this purpose are TT (Tensor Train)and HT (Hierarchical Tucker), both are the results of one and same schemefor the reduction of dimensionality. These decompositions are frequentlyapplied to tensors after some, often ultimate, quantization of the originaldimensions. This maximizes the number of modes and makes the number of countsat each mode minimal possible, e.g. 2. We consider how multilevel matrices become tensor trains with the use of the Kronecker-product operation. Then we show how new wavelet transforms arise in the construction oftensor trains when forcing the TT ranks be limited. We also discuss someexamples and perspectives for applications.  In the end, we present an ambitious research goal of design of fast tabulationprocedures using a new interpolation instrument based a TT generalization of the skeleton (dyadic) decomposition from matrices to tensors. REFERENCES [1] I.Oseledets, E. Tyrtyshnikov, TT-cross approximation for multidimensional arrays, Linear Algebra Appl., 432 (2010), pp. 70-88. [2] I. Oseledets, E. Tyrtyshnikov, Algebraic wavelet transform via quantics tensor train decomposition, SIAM J. Sci. Comp., vol. 31, no. 3 (2011), pp. 1315-1328. [3] I. Oseledets, E. Tyrtyshnikov, N. Zamarashkin, Tensor-train ranks for matrices and their inverses, Comput. Meth. Appl. Math., Vol. 11, No. 3 (2011), pp. 375-384.

Laurent Sorber, State of the Art Overview of Algorithms for  Tensor Decompositions

(slides)

The canonical polyadic and rank-(Lr,Lr,1) block term decomposition (CPD and BTD, respectively) are two closely related tensor decompositions. The CPD is an important tool in psychometrics, chemometrics, neuroscience and data mining, while the rank-(Lr,Lr,1) BTD is an emerging decomposition in signal processing and, recently, blind source separation. We present a decomposition that generalizes these two and give an overview of the state of the art in algorithms for these decompositions, including their drawbacks and benefits. Among these algorithms are recently developed (complex) optimization-based methods such as limited-memory BFGS and matrix-free nonlinear least squares techniques. In the latter, we exploit the structure of the Jacobian's Gramian by means of efficient expressions for its matrix-vector product. Combined with an effective preconditioner, numerical experiments confirm that these methods are among the most efficient and robust currently available for computing the CPD, rank-(Lr,Lr,1) BTD and their generalized decomposition.

Nick Vannieuwenhoven, A New Truncation Strategy for the Higher-order Singular Value Decomposition

(slides)

We present an alternative strategy to truncate the higher-order singular value decomposition (T-HOSVD), called the sequentially truncated HOSVD (ST-HOSVD). It requires less operations to compute and often improves the approximation error with respect to the T-HOSVD. Whenever a tensor is truncated to its multilinear rank, the basis computed by the T-HOSVD and ST-HOSVD coincide. In one experiment we performed, the results of a numerical simulation of a partial differential equation were compressed by T-HOSVD and ST-HOSVD. At the same truncation rank, the approximation errors were similar, and the bases computed by both algorithms were indistinguishable. The execution time, on the other hand, was reduced from 2h45 for T-HOSVD to one minute for ST-HOSVD, representing a speedup of 133. In another experiment, we compare T-HOSVD and ST-HOSVD for constructing a multilinear model for classifying handwritten digits.

Boris N. Khoromskij, Efficient Numerical Approximation of Multi-dimensional PDEs in Quantized Tensor Spaces

(slides)

Modern methods of tensor-product approximation by separation of variables allow an efﬁcient lowparametric calculus of functions and operators in higher dimensions. Most common separable representations combine the canonical, Tucker, tensor train (TT) and the quantized-TT (QTT) decompositions. The QTT tensor format makes it possible to represent (approximate) the multivariate functions, operators and dynamical systems in the quantized tensor spaces with the log-volume complexity scaling with respect to the size of full tensor [2]. This opens the way to the profound numerical simulation of high-dimensional PDEs getting rid of the “curse of dimensionality” and rigorous restrictions on the grid size. The numerical efﬁciency is justiﬁed by the remarkable QTT-approximation properties proven for the wide class of functions and operators [2, 3]. We focus on the approximation and complexity results in the quantized tensor formats applied to the solution of d-dimensional elliptic and parabolic equations [1] - [5]. Numerical tests indicate the logarithmic computational complexity of the QTT tensor approximation for the parametric elliptic PDEs, in computational quantum chemistry and in the multi-dimensional dynamics.

http://personal-homepages.mis.mpg.de/bokh

References

[1] I.V. Gavrilyuk, and B.N. Khoromskij. Quantized-TT-Cayley transform to compute dynamics and spectrum of high-dimensional Hamiltonians. Comp. Meth. in Applied Math., v.11 (2011), No. 3, 273-290.

[2] B.N. Khoromskij. O(d log N)-Quantics Approximation of N-d Tensors in High-Dimensional Numerical Modeling. J. Constr. Approx. v. 34(2), 257-289 (2011).

[3] B.N. Khoromskij. Tensors-structured Numerical Methods in Scientiﬁc Computing: Survey on Recent Advances. Chemometrics and Intellingent Laboratory Systems, 110 (2012), 1-19.

[4] B.N. Khoromskij, and Ch. Schwab. Tensor-Structured Galerkin Approximation of Parametric and Stochastic Elliptic PDEs. SIAM J. Sci. Comp., 33(1), 2011, 1-25.

[5] B.N. Khoromskij, and I. Oseledets. DMRG+QTT approach to the computation of ground state for the molecular Schrodinger operator. ¨ Preprint 68/2010, MPI MiS, Leipzig 2010 (Numer. Math., submitted).

Ignat Domanov, On the Uniqueness of the Canonical Polyadic Decomposition: New Sufficient Conditions Based on Properties of Khatri-Rao Products of Compound Matrices

(slides)

The decomposition of a tensor in a minimal linear combination of rank-1 tensors is called its Canonical Polyadic Decomposition (CPD). The decomposition is also known as Canonical / Parallel Factor Decomposition (CANDECOMP/PARAFAC). The most well-known result concerning CPD uniqueness is due to J. Kruskal: let A, B, C be the factor matrices in a CPD of a tensor T, and let kA+kB+kC≥2R+2,where kA, kB, kC denote the k-rank of A, B, C, respectively, then the CPD of T is unique. Other uniqueness conditions are due to T. Jiang and N.D. Sidiropoulos, and L. De Lathauwer. These conditions are on one hand more restrictive than Kruskal’s in the sense that one of the factor matrices, say C, needs to have full column rank, which implies that the rank R is bounded by the number of rows of C, i.e., by the third tensor dimension. On the other hand, the results are an order of magnitude more relaxed than Kruskal’s in terms of the conditions on A and B. The latter may be expressed as conditions on the Khatri-Rao product of the second compound matrices of A and B. In this talk we establish a theory for CPD uniqueness by working from the Jiang-Sidiropoulos-De Lathauwer results towards Kruskal’s result, thereby extending work of A. Stegeman. We relax the condition on C, no longer requiring that it has full column rank, while making the conditions on A and B more restrictive. The latter are conditions on the Khatri-Rao product of m-th compound matrices of A and B, where m > 2. In the other direction, we present a condition on A and B that ismore relaxed than the least restrictive Jiang-Sidiropoulos condition. Part of the study leads to new Kruskal-type conditions that are more relaxed than the original. For instance, we explain that, if kA+rB+rC≥2R+2, and rA+kB+rC≥2R+2, and rA+rB+kC≥2R+2, where rA, rB, rC denote the rank of A, B, C, respectively, then the CPD of T is unique. We discuss both deterministic and generic conditions. We show INDSCALvariants. We also present results concerning uniqueness of one factormatrix, regardless of overall uniqueness

Mikael Sorensen, Canonical Polyadic Decomposition with  Orthogonality Constraints

Canonical Polyadic Decomposition (CPD) of a higher-order tensor is an important tool in mathematical engineering. In several signal processing  applications  at least one of the matrix factors is constrained to be column-wise orthonormal.  If  the   tensor admits a CPD where one of the factors is constrained to be column-wise orthonormal, then it can be shown that it admits an algebraic solution under very mild conditions. However, in practical signal processing applications noise is always present and consequently  only a low-rank tensor approximation  is of interest.  Contrary to the unconstrained case, a  low-rank approximation  of a tensor by an orthogonality constrainted CPD always exists.  We also explain  that by taking the this constraint in account, more efficient and noise robust numerical algorithms for the computation of the constrained CPD can be obtained.  In particular, orthogonality-constrained versions of the  CPD  methods based on  simultaneous  matrix  diagonalization  and  alternating least squares  are presented.

Toon Van Waterschoot, Sensing Wave Fields in a Finite Element Framewor

(slides)

In this talk, we present a recently developed framework for sensing wave fields through the combination of data-based and model-based information. We consider two distinct ways of collecting wave field information, either by recording data measurements obtained from spatially distributed ensors or by assuming a physical field model in the form of a partial differential equation. We will then show how these two information sources can be jointly discretized in space and time by constructing an appropriate spatiotemporal sampling grid and employing the finite element method. This iscretization leads to a high-dimensional yet highly sparse and structured wave field representation, hich can then be used for a variety of applications: the estimation of static and dynamic wave fields, the inverse problem of source recovery, the localization of point sources, the identification of a wave propagation model, etc. A crucial issue in all of these applications is the possibility to formulate the related parameter estimation problem as a large-scale convex optimization problem. Furthermore, the sparse and structured wave field representation allows for an efficient distributed optimization approach, which is useful in applications featuring networked devices with local sensing, processing, and communication capabilities. Short link

Wed 11 - Wed 11 Apr-12 ERC A-DATADRIVE-B Start-up meeting
ESAT 01.60
9:30 am-4:30 pm

April 11, 2012
ESAT 01.60

- 09:30-09:40 Welcome newcomers

Johan Suykens

- 10:00-10:30 "Structured Sparsity, Overlap Norms and Optimization"
Andreas Argyriou

- 10:30-10:50 "Kernels and Tensors for Structured Data Modelling"
Marco Signoretto

- 10:50-11:10 break

- 11:10-11:40 "Continuous Piecewise Linear Identification and Optimization"
Xiaolin Huang

- 11:40-12:00 "LS-SVM Approach for Solving Linear Descriptor Systems"
Siamak Mehrkanoon

- 12:00-12:20 "Multiple Kernel Learning in Biomedical Applications"
Vilen Jumutc

- 12:20 - 14:00 lunch

- 14:00-14:20 "A Semi-Supervised Formulation to Binary Kernel Spectral Clustering"
Carlos Alzate

- 14:20-14:40 "Towards Sparseness in Generative Models"
Gervasio Puertas

- 14:40-15:00 "Kernel Spectral Clustering on Time series, and Model Selection for KPCA"
Carolina Varon

- 15:00-15:30 break

- 15:30-15:50 "Joint Regression and Linear Combination of Time Series for Optimal Prediction"
Dries Geebelen

- 15:50-16:10 "A Kernel-based Approach to Community Detection"
Rocco Langone

- 16:10-16:30 "Load Forecasting in a Multivariate Meta-learning System"
Marin Matijas

Thu 15 - Thu 15 Mar-12 Quarterly Meeting of OPTEC WG 3
CIT 91.45
9:00 am-11:00 am
Quarterly WG3 meeting.

Preliminary agenda:
- 09.00-09.10: welcome and agenda
- 09.10-09.30: update by members
- 09.30-10.00: Presentation by Dries Telen: "Including robustness and
multi-objective optimization in optimal experiment design of dynamic
(bio)chemical processes"
- 10.00-10.15: break
- 10.15-10.45: discussion and planning of next activities
- 10.45-11.00: aob

Fri 24 - Fri 24 Feb-12 KU Leuven Seminars on Optimization in Engineering - (WG2) Carlos Alzate, Siamak Mehrkanoon
ESAT 00.62
2:00 pm
14:00 "A Semi-Supervised Formulation to Binary Kernel Spectral Clustering"
Carlos Alzate (KU Leuven, ESAT-SCD)

14:30 "Parameter estimation in state-space models using LS-SVM"
Siamak Mehrkanoon (KU Leuven, ESAT-SCD)

Fri 17 - Fri 17 Feb-12 Workshop and Franqui Lecture of Yurii Nesterov
Liège
9:30 am-5:00 pm
"Workshop and Franqui Lecture of Yurii Nesterov"

February, 17 2012

9:30 - 12:30:
• Etienne de Klerk (Tilburg University)
• Alexandre d'Aspremont (École Polytechnique, Paris)
• Michel Baes (ETH Zürich)
15:00
Yurii Nesterov: Algorithmic Challenges in Optimization
Mathematical Point of View

Register at:
http://www.montefiore.ulg.ac.be/~francqui/

Franqui main course by Y. Nesterov:

February, 24 2012 - March, 23 2012, 10:30 - 12:30
Institut Montefiore, building B28, room R7
• 24/02: Intrinsic complexity of Black-Box Optimization
• 02/03: Looking into the Black Box (Structural Optimization)
• 09/03: Huge-scale optimization problems
• 16/03: Nonlinear analysis of combinatorial problems
• 23/03: Algorithmic models of human behavior

Thu 16 - Thu 16 Feb-12 Quarterly Meeting of OPTEC WG 4
Mechanical Engineering, Baekeland room
2:00 pm-4:00 pm

Quarterly Meeting of OPTEC WG 4 on Shape and Topology Optimization

Proposed agenda

- Continue discussion on topology optimization: how to constraint complexity of the solution in a rigorous setting? (30 min – hope the topology adepts are willing to take the lead once more – to all others, please refresh your memory, cf. presentation in attachment)

- A PDE-constrained optimization problem – problem presentation by Joel Anderson (to be confirmed– 20 min)

- coffee break + informal discussion (15 min)

- Date for next meeting + agenda (10 min)

- Brainstorm on connection between applied math and application: try to define simple problems for CS (connect to exercises? Simple paper?). Setup discussion on possible joint project proposals. (2 times 15 min in smaller groups)

Wed 15 - Thu 24 May-12 Ph.D. course on convex optimization
ESAT
10:30 am-12:30 pm
"Ph.D. course on convex optimization"

Goele Pipeleers and Quoc Tran Dinh

Course Summary:
This course concentrates on recognizing and solving convex optimization problems that arise in engineering, and covers the following topics:
- Convex sets, functions, and optimization problems.
- Optimality conditions, duality theory, theorems of alternative, and applications.
- Interior-point methods.
- Applications to signal processing, control, digital and analog circuit design, computational geometry, statistics, and mechanical engineering.

[PDF]

Wed 15 - Wed 15 Feb-12 K.U. Leuven Seminars on Optimization in Engineering - Martin Mönnigmann
ESAT 00.62
4:30 pm-5:30 pm
"Fast explicit model predictive control"

M. Mönnigmann
Automatic Control and Systems Theory, Ruhr-Universität Bochum, Germany

Model predictive control (MPC) is an established method for the control of constrained multivariable systems. New theoretical insights, improved and tailored optimization algorithms, and the ever growing performance of hardware have helped MPC advance to higher and higher sampling frequencies. For linear systems and sampling times below a millisecond, explicit MPC (EMPC) methods are an interesting alternative. In EMPC it is no longer necessary to solve a receding horizon optimal control problem online. Instead, an analytical expression for the MPC control law can be found by solving a parametric optimization problem offline. EMPC can, however, still only be applied to very simple problems with short horizons for two reasons: The parametric optimization problem is more complex than its non-parametric receding horizon counterpart. Secondly, the expression for the explicit control law u(x) may grow so large that a naive online evaluation of u(x) takes as much time as solving the receding horizon optimal control problem online. The talk summarizes recent progress with respect to both obstacles, the offline calculation of explicit control laws, and their fast online evaluation. Specifically, a simple new approaches to the fast evaluation of EMPC control laws results in online evaluation times on the order of 10ns. This approach does not require a CPU, but it can be implemented on low-cost, compact hardware with low power consumption such as programmable gate arrays. Progress in the fast evaluation of EMPC control laws has triggered the development of new approaches to solving the offline optimization problem. To this end, a new approach is suggested that avoids the state space exploration common to most existing methods.

Thu 9 - Thu 9 Feb-12 SISTA Seminar - Kris De Brabanter
ESAT 00.62
2:00 pm
"Properties of Linear Smoothers"

Kris De Brabanter (K.U. Leuven, ESAT-SCD)

In this talk we illustrate some properties of a large class of modeling techniques called linear smoothers. Examples of linear smoothers include Nadaraya-Watson kernel regression, local polynomial regression, LS-SVM, splines, wavelets, orthogonal series,... In case of linear smoothers it is relatively easy to study basic properties such as bias, variance and mean squared error. As a direct consequence confidence and prediction intervals can be easily constructed for regression and classification. In a second part of this talk we will focus the attention on recent asymptotic results regarding this class of smoothers.  We develop conditions for asymptotic normality and asymptotic negligibility of linear smoothers. Further, we give uniform and local Berry-Esseen bounds for linear smoothers and establish rates of convergence for the kernel regression estimator. Finally, we establish exponential bounds on the tail distribution and present  a variant of the weak and strong law of large numbers for linear smoothers.

Tue 7 - Tue 7 Feb-12 OPTEC MANET Workshop on Tensors and Large Scale Optimization
ESAT 00.62
2:00 pm-6:00 pm
"OPTEC MANET Workshop on Tensors and Large Scale Optimization"

This internal workshop brings together OPTEC professors, postdocs, and a few PhD students that are interested in Tensors and in Large Scale Optimization Algorithms. Each of the present professors gives a 10 minute statement on large scale optimization problems and algorithms in his/her group.
The workshop is related to OPTEC's Special Interest Groups on "Tensors and Optimization" (coordinated by Lieven De Lathauwer) and "Distributed and Parallel Methods for Optimization" (coordinated by Toon van Waterschoot), as well as to the GOA project MANET. The workshop will reserve 50% of its time for discussions. Aim is to find synergies between the groups.

Program so far:

12:00 Joint Alma lunch for all who like to join

14:00 Start and introduction
14:10 Lieven De Lathauwer: Tensor Optimization Problems
14:20 Discussion
14:30 Johan Suykens: Large Scale Optimization in Machine Learning
14:40 Discussion
14:50 Moritz Diehl: Distributed Optimization Algorithms
15:00 Discussion
15:10 Toon van Waterschoot: Distributed Optimization Problems in Signal Processing
15:20 Discussion

15:30 Coffee break

16:00 Sabine Van Huffel: Tensor Optimization in Biomedical Applications
16:10 Discussion
16:20 Alexander Bertrand: Consensus-based Distributed Total Least Squares Estimation in Ad-hoc Networks
16:35 Laurent Sorber: Efficient Algorithms for Tensor Decompositions
16:50 Sam Weckx: Distributed Optimization in Smart Grid Applications
17:05 Marco Signoretto: Learning Tensor-based Models with Structure-inducing Penalties
17:20 future planning of cooperations and wrap up
18:00 end

Workshop Organizers: Lieven De Lathauwer, Toon van Waterschoot, Moritz Diehl

Tue 7 - Tue 7 Feb-12 K.U. Leuven Seminars on Optimization in Engineering - Tobias Lindstrøm Jensen
ESAT 00.62
11:00 am-12:00 pm
"A Lower Complexity Bound for l1-regularized Least-squares Problems using a Certain Class of Algorithms"

Tobias Lindstrøm Jensen
Deparment of Electronic Systems at Aalborg University

The l1-regularized least-squares problem have received broad attention the last couple of years. The result is numerous approaches for reliable large-scale solvers which combines both well known methods and recently developed techniques for efficient computations. We define a class of algorithms which is not as restrictive as classic black-box algorithms and hence includes most of the recently proposed methods. We show how to obtain a worst-case convergence rate for all these methods.

Tue 31 - Tue 31 Jan-12 K.U. Leuven Seminars on Optimization in Engineering - Andreas Potschka
ESAT 00.62
4:00 pm-5:00 pm
"Optimization problems with time-periodic PDE constraints"

Andreas Potschka
Interdisciplinary Center for Scientific Computing (IWR)Heidelberg University

Optimization problems with time-periodic parabolic PDE constraints arise in important application areas, e.g., in chemical engineering. The resulting nonlinear dynamical optimization problems are difficult, especially because they feature free initial values. We present a novel direct numerical optimization method based on inexact Sequential Quadratic Programming (SQP) with a two-grid Newton-Picard preconditioned Linear Iterative Splitting Approach (LISA) for the quadratic subproblems (QPs). The method features fast linear convergence that is independent of the degrees of freedom for the fine spatial discretization grid. We demonstrate how the arising large-scale QPs can be solved efficiently via intelligent structure exploitation. Moreover, we address issues of affine-invariant globalization of convergence and discuss local convergence of LISA-SQP in the framework of Bock's kappa-Theory. We present how novel a-posteriori kappa-estimators can be used to control the local rate of convergence by adaptively choosing the coarse grid from a given hierarchy of grid levels. Finally, we illustrate the performance of the method by numerical results for three application problems ranging from an academic model problem to a real-world periodic adsorption process.

Fri 27 - Fri 27 Jan-12 Recent advances in applied model predictive control - Alberto Bemporad
FMTC, Celestijnenlaan 300D, 3001 Heverlee
1:30 pm-3:30 pm
"Recent advances in applied model predictive control"

Professor of Control Systems at the IMT Institute for Advanced Studies Lucca, Italy

This talk is part of a workshop of the Lecopro project of IWT,
http://www.lecopro.org/midterm_workshop.html
The talk is particularly recommended to OPTEC members in WG1.

Tue 24 - Tue 24 Jan-12 K.U. Leuven Seminars on Optimization in Engineering - WG2 SIGs
ESAT 00.62
2:00 pm-3:00 pm

Presentation of OPTEC-WG2 related special interest groups:

14:00 Mathematical Statistics in Optimization (De Brabanter J., De Brabanter K.)

14:30 Tensors (De Lathauwer L., Van Barel M.)

Mon 16 - Mon 16 Jan-12 WG3 Seminar - Kris De Brabanter
CIT 91.45
10:00 am-11:00 am

Presentation of Special Interest Group on "Mathematical Statistics in
Optimization" by Kris De Brabanter

Theme: How to detect correlation?

Thu 12 - Thu 12 Jan-12 Quarterly Meeting of OPTEC WG 1 on Dynamic and Embedded Optimization
ESAT 00.62
9:30 am-12:15 pm
"Quarterly Meeting of OPTEC Working Group 1 on Dynamic and Embedded Optimization"

Agenda

09:30 Introduction
09:45 Mattia Vallerio: "Towards enhanced weight selection for (N)MPC via multi-objective optimisation".

10:15 Coffee Break
10:45 Joel Andersson: "Dynamic optimization of a combined cycle power plant".

11:30 Discussion
12:15 End

Fri 16 - Fri 16 Dec-11 Doctoral Presentation - Marco Signoretto
Auditorium of Arenberg Castle, Kasteelpark Arenberg 1, 3001 Heverlee (Leuven)
2:00 pm
Kernels and Tensors for Structured Data Modelling

Marco Signoretto (K.U. Leuven, ESAT-SCD)

Abstract
A key ingredient to improve the generalization of machine learning algorithms is to convey prior information, either by choosing appropriate input representations or by tailored regularization schemes. This becomes of paramount importance in all the applications where the number of available observations for training is limited. In many of such cases, data are structured and can be conveniently represented as higher order arrays (tensors). The scope of this thesis is the development of learning algorithms that exploit the structural information of these arrays to improve generalization. This is achieved by combining tensor-based methods with kernels, convex optimization, sparsity and statistical learning principles.

As a first contribution we present a parametric framework based on convex optimization and spectral regularization. We give a mathematical characterization of spectral penalties for tensors and analyze a unifying class of convex optimization problems for which we present a new, provably convergent and scalable template algorithm. We then specialize this class of problems to perform learning both in a transductive as well as in an inductive setting. In the transductive case one has an input data tensor with missing features and, possibly, a partially observed matrix of labels. The goal is to both infer the missing input features as well as predict the missing labels. For induction, the goal is to determine a parametric model for each learning task to be used for out of sample prediction. Each training pair consists of a multidimensional array and a set of labels each of which corresponds to related but distinct tasks. As a by-product of using a tensor-based formalism, our approach enables one to tackle multiple tasks simultaneously in a natural way. Empirical studies demonstrate the merits of the proposed methods.

Parametric tensor-based techniques present a number of advantages; in particular, they often lead to interpretable models which is a desirable feature in a number of applications of interest. However they constitute a somewhat restricted class that might suffer from limited predictive power. A second contribution of this thesis is to go beyond this limitation by introducing nonparametric tensor-based models. To this end we present two different ideas. The first approach is based on an explicit multi-way feature representation. A main drawback is that estimation within this feature space results into non-convex and non-scalable problems. The second approach fits into the same primal-dual framework underlying SVM-like algorithms and allows the efficient estimation of nonparametric tensor-based models. Although specialized kernels exist for certain classes of structured data, no existing approach exploits the structure of tensorial representations. We go beyond this limitation by proposing a class of tensorial kernels that links to the multilinear singular value decomposition (MLSVD) and study properties of the proposed similarity measure.

Many objects of interest, such as videos and colored images, admit a natural tensorial representation. Additionally, tensor representations naturally result from the experiments performed in a number of fields. On top of this, there are cases where one can explicitly carry on tensor transformations with the purpose of exploiting the spectral content of these new representations. We show that one of such transformations can be used for learning when input data are multivariate time series. Contrary to existing approaches, the resulting procedure does not require the (often nontrivial) blind identification of generative models. The approach is illustrated on a brain decoding task where the direction, either left of right, towards where the subject modulates attention is predicted from magnetoencephalography (MEG) signals.

Promotors: Prof. J. Suykens, Prof. J. Vandewalle, Prof. L. De Lathauwer

Wed 30 - Wed 30 Nov-11 K.U.Leuven Robotics Event
Robotics Research Laboratory (room 01.39)
2:00 pm-7:00 pm
K.U.Leuven Robotics Event
Robotics Research Laboratory (room 01.39)
Katholieke Universiteit Leuven, Dept. of Mechanical Engineering
Celestijnenlaan 300b, B-3001 Heverlee
The Department of Mechanical Engineering of the K.U.Leuven introduces its robotics related research to the general public.
The Robotics and MECO research groups open the doors to their remodeled lab and illustrate their expertise,
covering autonomous compliant motion, robot-assisted surgery, mobile robotics, advanced motion control…, on real-life demos.
Please register by November 15, 2011 on
www.mech.kuleuven.be/roboticsevent2011

Tue 29 - Tue 29 Nov-11 DYSCO day and lecture L. Biegler on Dynamic Real-Time Optimization
Aud 2de Hoofdwet
10:00 am
DYSCO day and lecture L. Biegler on Dynamic Real-Time Optimization

Lorenz T. Biegler, Carnegie Mellon University, USA:
Direct Transcription Strategies for Dynamic Real-time Optimization

http://sites.uclouvain.be/dysco/StudyDays/2011-november

Mon 28 - Mon 28 Nov-11 Simon Stevin Lecture on Optimization in Engineering - Michel Perrier
Auditorium De Molen - Arenberg Castle
4:00 pm-5:30 pm
21st Simon Stevin Lecture on Optimization in Engineering, and 3rd Chemical Engineering Department Lecture (International Year of Chemistry)

"Online Optimization of Biochemical Processes"

Michel Perrier
Department of Chemical Engineering at Polytechnique Montréal

flyer, poster

Abstract

Optimization methods have been widely used in biology related fields with great success for example in systems biology, metabolic engineering, biochemical engineering amongst others. Most of the work has been focused on developing mathematical models trying to describe the steady-state and dynamical behaviour obtained from the observation of living organisms at the scale of single cell up to biochemical reactors. Optimization is then performed typically offline to obtain a better understanding of the organism under study from a design or operational standpoint.  Comparatively, fewer studies have been performed with the objective of optimizing a specified performance index on-line.

There are many problems specifically associated with online optimization compared to off-line. Model mismatch, model uncertainty, unmeasured disturbances have to be taken into account to make sure to optimize the actual system and not the model. The number of degrees of freedom is also much lower. Methods on how to address these problems will be given by studying several applications to biochemical systems with a focus on bioreactors. Perspectives for future research will be outlined.

Biographical Information

Prof. Michel Perrier is presently Head of the Department of Chemical Engineering at Polytechnique Montréal since Dec. 2009. He obtained his Ph.D. from McGill University. Before joining Polytechnique  as a professor in 1993, he worked in industry as a control engineer for a period of seven years at Shell Canada, the Pulp and Paper Research Institute of Canada and the Biotechnology Research Institute in Montréal. He has been a visiting professor at the Centre for Integrated Dynamics and Control in the department of Electrical Engineering at the University of Newcastle in Australia in 2001, at the Centre for Systems Engineering and Applied Mechanics at the Université Catholique de Louvain in Belgium in 2002, and at University Polytechnic of Catalunya in Spain in 2008-2009. His research interests are in the field of dynamics, control and optimization of biotechnological processes. He has been vice-chair and then chair of the technical committee on Biosystems and Bioprocesses of the International Federation of Automatic Control (IFAC) from 2001 to 2007. In 2008, he was awarded an honorary doctorate from Faculté Polytechnique de Mons in Belgium and in 2009 the D.G. Fisher Award from the Systems and Control Division of the Canadian Society for Chemical Engineering. In June 2010, he was inducted Fellow of the Canadian Academy of Engineering.

The "Simon Stevin Lecture Series on Optimization in Engineering" is set up in order to promote optimization in engineering. For this aim, every quarter of the year an outstanding international scholar is invited to report on latest progress in the development of optimization algorithms and their applications in engineering.
Simon Stevin (1548-1620) was a Flemish mathematician and engineer. Among other, he helped to advance the use of decimal fractions, was the first to explain the tides by the attraction of the moon, and discovered the hydrostatic paradox. He made numerous inventions, among them a wind propelled carriage with sails, the "land yacht", which once impressed Prince Maurice of Orange as it moved faster than horses, in around 1600 on the beach between Scheveningen and Petten. Simon Stevin was fond of promoting the use of science in daily life and in craftmanship, and translated various mathematical terms into dutch. Among other, he introduced the dutch word for mathematics, "wiskunde".

Wed 16 - Wed 16 Nov-11 ERC A-DATADRIVE-B Intro
ESAT AUD A
4:00 pm

http://www.kuleuven.be/research/erc/suykens.html
Johan Suykens (K.U.Leuven, ESAT-SCD)

16:30 Drink (Room 00.62)

Thu 6 - Thu 6 Oct-11 Doctoral Presentation - Hans Joachim Ferreau
Aula van de tweede hoofdwet (Thermotechnisch Instituut)
2:00 pm
"Model Predictive Control Algorithms for Applications with Millisecond Timescales"
Hans Joachim Ferreau

Promotors: Moritz Diehl, Joos Vandewalle

Abstract:

The last three decades have seen a rapidly increasing number of
applications where model predictive control (MPC) led to better control
performance than more traditional approaches. This thesis aims at
lowering the practical burden of applying fast MPC algorithms in the
real-world. To this aim, it contributes two software packages, which are
released as open-source code in order to stimulate their widespread use.
Both packages implement previously published methods but enrich them
with a number of new theoretical and algorithmic ideas.

The first part of this thesis focusses on efficiently solving quadratic
programs (QPs) as arising in linear MPC problems. To this end, it
reviews the author's previous work on developing an online active set
strategy to exploit the parametric nature of these QPs. This strategy is
extended with ideas for initialising the solution procedure and treating
QPs with semi-definite Hessian matrices. The software package qpOASES
implements the online active set strategy and its extensions together
with a number of tailored solution variants for special QP formulations.
It offers interfaces to third-party software like Matlab/Simulink and
has been successfully used in a number of academic real-world MPC
applications. Moreover, two industrial applications of qpOASES--dealing
with emission control of integral gas engines and feasibility management
for MPC in the process industry--are described. These industrial case
studies also led to further theoretical ideas, namely the use of MPC
with an asymmetric cost function and a novel method for handling
infeasible QPs based on the online active set strategy.

Toolkit, a new software environment and algorithm collection for
automatic control and dynamic optimisation. It has been designed for
setting-up nonlinear optimal control and MPC problems in a user-friendly
way and solving them efficiently. In particular, the ACADO Toolkit
implements two algorithmic variants of the real-time iteration scheme: a
Gauss-Newton approach for nonlinear MPC formulations involving a
tracking objective function as well as an exact Hessian approach for
tackling time-optimal formulations. The underlying QP subproblems are
solved by means of the online active set strategy. The ACADO Toolkit
features an intuitive symbolic syntax for formulating MPC problems,
which offers a couple of advantageous possibilities. Most importantly,
it allows the user to automatically generate optimised, highly efficient
C code that is tailored to each respective MPC problem formulation.
Numerical results show that the exported code exhibits a promising
computational performance allowing application of nonlinear MPC to
non-trivial processes at kilohertz sampling rates.

Thu 22 - Thu 22 Sep-11 K.U. Leuven Seminars on Optimization in Engineering - Ryota Tomioka
ESAT 00.62
3:00 pm

"Convex Tensor Decomposition with Performance Guarantee"
Ryota Tomioka (University of Tokyo)

Abstract:
We analyze the statistical performance of a recently proposed convex tensor
decomposition algorithm. Conventionally tensor decomposition has been
formulated as non-convex optimization problems, which hindered the
analysis of their performance. We show under some conditions that
both the performance of noisy tensor decomposition and tensor completion
can be predicted by the quantity we call the normalized rank.
Numerical experiments show that our theory can precisely predict
the scaling behavior in practice. The current analysis naturally
extends the analysis of convex low-rank matrix estimation to tensors.
We also discuss some limitations of our theory, open issues, and
possible extensions.

(OPTEC-WG2: data-driven modelling)

Fri 16 - Fri 16 Sep-11 Doctoral Presentation - Niels Haverbeke
ESAT AUD A
10:00 am

"Efficient numerical methods for moving horizon estimation"

Niels Haverbeke (K.U. Leuven, ESAT-SCD)

Abstract:

In model-based predictive control strategies, accurate estimates of the current state and model parameters are required in order to predict the future system behavior for a given control realization. One particularly powerful approach for constrained nonlinear state estimation is Moving Horizon Estimation (MHE). In MHE past measurements are reconciled with the model response by optimizing states and parameters over a finite past horizon. The basic strategy is to use a moving window of data such that the size of the estimation problem is bounded by looking at only a subset of the available data and summarizing older data in one initial condition term. This also establishes an exponential forgetting of past data which is useful for time-varying dynamics.

Compared to other state estimation approaches, MHE offers many advantages following from its formulation as a dynamic optimization problem. Inequality constraints on the variables (states, parameters, disturbances) can be included in a natural way and the nonlinear model equation is directly imposed over the horizon length. Empirical studies show that MHE can outperform other estimation approaches in terms of accuracy and robustness. In addition to these well-known advantages, the framework of MHE allows for formulations different from the traditional (weighted) least-squares formulation.

The greatest impediment to a widespread acceptance of MHE for real-time applications is still its associated computational complexity. Despite tremendous advances in numerical computing and Moore’s law, optimization-based estimation algorithms are still primarily applied to slow processes. In this work, we present fast structure-exploiting algorithms which use robust and efficient numerical methods and we demonstrate the increased performance and flexibility of nonlinear constrained MHE. MHE problems are typically solved by general purpose (sparse) optimization algorithms. Thereby, the symmetry and structure inherent in the problems are not fully exploited. In addition, the arrival cost is typically updated by running a (Extended) Kalman filter recursion in parallel while the final estimate covariance is computed from the derivative information. In this thesis, Riccati based methods are derived which effectively exploit the inherent symmetry and structure and yield the arrival cost update and final estimate covariance as a natural outcome of the solution process. The primary emphasis is on the robustness of the methods which is achieved by orthogonal transformations.

When constraints are imposed, the resulting quadratic programming (QP) problems can be solved by active-set or interior-point methods. We derive modified Riccati recursions for interior-point MHE and show that square-root recursions are recommended in this context because of numerical conditioning. We develop an active-set method which uses the unconstrained solution obtained from Riccati recursions and employs a Schur complement technique to project onto the reduced space of active constraints. The method involves non-negativity constrained QPs for which an efficient gradient projection method is proposed. We implement the algorithms in efficient C code and demonstrate that MHE is applicable to fast systems.

These QP methods are at the core of solution methods for general convex and nonlinear MHE as is demonstrated. Convex formulations are investigated for robustness to outliers and abrupt parameter changes. Furthermore, the methods are embedded in a Sequential Quadratic Programming strategy for nonlinear MHE. One application has been of particular interest during this doctoral research: estimation and predictive control of blood-glucose at the Intensive Care Unit (ICU). For this application reliability and robustness of the estimates as well as of the numerical implementations are crucial. We evaluate an MHE based MPC control strategy and show its potential for this application.

Promotors: Prof. B. De Moor, Prof. M. Diehl

Thu 15 - Thu 15 Sep-11 Doctoral Presentation - Xinhai Liu
ESAT AUD A
5:00 pm
"Learning from multi-view data: clustering algorithm and text mining application"

Xinhai Liu (K.U.Leuven, ESAT-SCD)

Abstract: The rapid development of information and computer technology
(ICT) in the last two decades has fundamentally changed almost every
discipline in science and engineering, transforming many fields from
data-poor to increasingly data-rich, and calling for innovative data
mining methods to conduct the related research. Meanwhile, as data
collection sources and channels continuously evolve, data can be
extracted from multiple information sources and observed by various
models. Therefore, learning from multi-view data has become a crucial
step in machine intelligence and knowledge discovery.

For the purpose of integrating and leveraging the mass amount of
multi-view data to obtain significant and complementary high-level
knowledge, this dissertation investigates learning from multi-view data
from two sides: clustering algorithm and text mining application. The
dissertation is organized into three parts. In the first part, we
analyze multi-view clustering from multilinear perspective and create
several novel multi-view clustering algorithms. In the second part, we
investigate text mining to extract multi-view heterogeneous data from a
large-scale publication database of Web of Science (WoS) and propose
several partitioning strategies for scientific mapping. In the third
part, we propose a novel strategy to derive knowledge from textual
information from a multi-view perspective. The theory, algorithm,
applications and software presented in this dissertation provide an
interesting perspective for clustering algorithms and text mining
applications. In addition, the obtained results are promising to be
applied and extended to many other relevant fields besides scientific
mapping and bioinformatics.

Promotor: Prof. B. De Moor

Wed 7 - Wed 7 Sep-11 Doctoral Presentation - Boris Houska
Aud. A
5:00 pm
"Robust Optimization of Dynamic Systems"

Boris Houska (K.U. Leuven, ESAT-SCD)

Abstract

This thesis is about robust optimization, a class of mathematical optimization problemswhich arise frequently in engineering applications, where unknown process parameters and
unpredictable external influences are present. Especially, if the uncertainty enters via a
nonlinear differential equation, the associated robust counterpart problems are challenging
to solve. The aim of this thesis is to develop computationally tractable formulations
together with efficient numerical algorithms for both: finite dimensional robust optimization
as well as robust optimal control problems.

The first part of the thesis concentrates on robust counterpart formulations which lead to
“min-max” or bilevel optimization problems. Here, the lower level maximization problem
must be solved globally in order to guarantee robustness with respect to constraints.
Concerning the upper level optimization problem, search routines for local minima are
required. We discuss special cases in which this type of bilevel problems can be solved
exactly as well as cases where suitable conservative approximation strategies have to be
applied in order to obtain numerically tractable formulations. One main contribution of this
thesis is the development of a tailored algorithm, the sequential convex bilevel programming
method, which exploits the particular structure of nonlinear min-max optimization problems.

The second part of the thesis concentrates on the robust optimization of nonlinear dynamic
systems. Here, the differential equation can be affected by both: unknown time-constant
parameters as well as time-varying uncertainties. We discuss set-theoretic methods for
uncertain optimal control problems which allow us to formulate robustness guarantees
with respect to state constraints. Algorithmic strategies are developed which solve the
corresponding robust optimal control problems in a conservative approximation. Moreover,
the methods are extended to open-loop controlled periodic systems, where additional
stability aspects have to be taken into account.

The third part is about the open-source optimal control software ACADO which is the basis
for all numerical results in this thesis. After explaining the main algorithmic concepts and
structure of this software, we elaborate on fast model predictive control implementations
for small scale dynamic system as well as on an inexact sequential quadratic programming
method for the optimization of large scale differential algebraic equations. Finally, the
performance of the algorithms in ACADO is tested with robust optimization and robust
optimal control problems which arise from various fields of engineering.

Tue 6 - Tue 6 Sep-11 WG3: Book reading meeting "Applied parameter estimation for chemical engineers"
CIT 91.45
10:00 am-12:00 pm
"WG3: Book reading meeting "Applied parameter estimation for chemical
engineers.""

WG3 has started a book reading group. Aim is to read together basic
books on parameter estimation to strengthen the knowledge. Hence,
everyone who is interested, can join.

Format:
There is one meeting of 2 hours every two weeks. Each time two persons
study one or two chapters and they give a tutorial presentation with
slides to the others (1 hour) who have read the chapters. Afterwards
there is 1 hour for discussion.

Book:
The first book that will be read is "Applied parameter estimation for
chemical engineers." Despite the title, the book provides a general
introduction and no chemical engineering background is needed.
Information on the book can be found at the WG3 page of the intranet of
the OPTEC website:
http://www.kuleuven.be/optec/internal/groups/WG3/137-WG3-Meeting-Presentations

Tue 7 - Tue 7 Jun-11 SISTA Seminar - Adrien Combaz
ESAT 01.60
10:00 am

"EEG based Brain Computer Interfaces for Communication"

Adrien Combaz (K.U. Leuven, Laboratory for Neuro- and Psychofysiology)

This seminar treats of Brain Computer Interface systems for spelling words based on the brain activity recorded via electroencephalography (EEG). It will present 2 systems based on different types of brain activity (P300 Event-Related Potential and Steady State Visually Evoked Potentials), describe studies performed with healthy and disabled subjects and discuss possible improvements.

Fri 27 - Fri 27 May-11 Doctoral Presentation - Julian Bonilla
Wolfspoort Auditorium, Schapenstraat 34, 3000 Leuven
5:00 pm
Structure and convexity exploitation of nonlinear chemical process modeling and estimation

Julian Bonilla

AbstractThe field of chemical process  estimation and control has been intensively explored in the last decades. However, novel applications, the demands  required by persevering safety regulations, tighten environmental standards, operating constraints and product quality specifications, originate more difficult and challenging situations. This creates the need of more sophisticated solutions than the ones that can be provided by traditional techniques alone. (i) Exploiting  the process model structure along with (ii) methods to deal efficiently with estimation and control problems are of paramount importance to reduce the computational load that new techniques often demand. This dissertation explores these two aspects for a particular class of first principles dynamic models. On the one hand, structure exploitation is studied through a widely used input-affine chemical process, namely distillation. A rigorous model is developed for a packed distillation column, leading to large scale differential-algebraic equations (DAEs). It is shown that these DAEs  can be  reduced by constraints differentiation and algebraic manipulation, preserving the physical meaning of the states in the representation. This kind of models exhibits high differentiation index, making its simulation impossible with off-the-shelf solvers. Hence, a simple procedure, based on the model Jacobian structural properties, is proposed in order to reduce the index of the model. Moreover, the reduced index DAEs  are cast such that sparse structures are obtained for simulation tasks, alleviating the computational load when solving the model.On the other hand, input/parameter-affine models are analyzed in the formulation of dynamic optimization problems (DOP). It is shown that a DOP using this kind of models, with convex cost and convex inequality constraints, can be approximated by a convex DOP. This approximation is performed by formulating a parametric DOP whose extremes correspond to the original nonconvex DOP and to a convex one.  The method is applied in the context of optimal control and  parameter estimation, such that a simple 2-step approach is proposed as an alternative to the solution of the original nonconvex DOP. In this form, the computational load involved in solving a parameterized DOP exactly,  is reduced by a simple 2-step convex optimization method that leads to a near optimal solution.

Promotors: Prof. J. Van Impe, Prof. B. De Moor
Co-promotor: Prof. M. Diehl

Thu 26 - Thu 26 May-11 SISTA Seminar - Kim Batselier
ESAT 00.57
2:00 pm
The Geometry of Polynomial Division

Kim Batselier (K.U. Leuven, ESAT-SCD)

Polynomial division is typically discussed in two ways. The first one is symbolic, where the division algorithm is the well-known polynomial long division from high school. The other way is using linear systems theory. Everybody is familiar with how univariate polynomial division corresponds with a deconvolution operation. These two viewpoints however do not generalize well to the multivariate case. In this seminar a new framework will be presented that allows to describe both univariate as multivariate polynomial division in a unified way. Since this framework is rooted in linear algebra a geometric interpretation will be given as well.

Tue 24 - Wed 25 May-11 Airborne Wind Energy Conference - AWEC2011
Leuven
9:00 am-5:00 pm
"Airborne Wind Energy Conference - AWEC2011"
May 24-25, 201, Leuven, Belgium

Conference webpage: http://www.awec2011.com

Organized by K.U. Leuven and its optimization in engineering center OPTEC,
working together with the Airborne Wind Energy Consortium and Ampyxpower.

Fri 20 - Fri 20 May-11 Doctoral Presentation - Fabian Ojeda
ESAT AUD A
2:30 pm
Kernel based Methods for Microarray and Mass Spectrometry Data Analysis

Fabian Ojeda (K.U. Leuven, ESAT-SCD)

Microarrays technologies allow to measure simultaneously the expression levels of thousands of genes contained in a biological sample. In the same way, mass spectrometry technologies generate large amounts of data measuring the activity of thousands of proteins. Common problems to these biological data sources involve the large number of variables (genes or proteins) compared to the number of samples, background noise and the presence of irrelevant variables, missing values and outliers. Regularization and kernel methods are mathematical techniques for linear and non-linear modeling, which are capable of handling high-dimensional data under a minimal set of assumptions. In this thesis, we apply and extend existing kernel methods for the analysis of high dimensional data sets from microarray and mass spectrometry technologies. Firstly, based on the structure of the Least-squares support vector machine (LS-SVM) methods, we propose an efficient algorithm to select relevant variables in the classification of microarray samples. This method updates the model parameters, thus allowing computations to scale up to thousands of variables. Secondly, we  develop regularized models for Mass Spectral Imaging (MSI) to help predict the anatomical categories of uncategorized regions on a mouse brain tissue data set. To overcome the lack of labeled tissue areas, we model the spatial proximity of neighboring spectra. Finally, we explore the use of kernel spectral clustering to group large number of genes based on their expression levels. In this approach, an small subset genes maximizing the sample entropy is first selected, and subsequently the clustering model is used to infer groups of the remaining genes.

Promotor: Prof. B. De Moor
Co-promotor: Prof. J. Suykens

Wed 27 - Wed 27 Apr-11 Doctoral Presentation - Kris De Brabanter
Auditorium of Arenberg Castle, Kasteelpark Arenberg 1, 3001 Heverlee (Leuven)
6:00 pm

"Least squares support vector regression with applications to large-scale data: a statistical approach"

Kris De Brabanter (K.U. Leuven, ESAT-SCD)

Nonparametric regression is a very popular tool for data analysis because these techniques impose few assumptions about the shape of the mean function. Hence, they are extremely flexible tools for uncovering nonlinear relationships between variables. A disadvantage of these methods is their computational complexity when considering large data sets. In order to reduce the complexity for least squares support vector machines (LS-SVM), we propose a method called Fixed-Size LS-SVM which is capable of handling large data set on standard personal computers.

We study the properties of the LS-SVM regression when relaxing the Gauss-Markov conditions. We propose a robust version of LS-SVM based on iterative reweighting with weights based on the distribution of the error variables. We show that the empirical maxbias of the proposed robust estimator increases slightly with the number of outliers in region and stays bounded right up to the breakdown point. We also establish three conditions to obtain a fully robust nonparametric estimator.

We investigate the consequences when the i.i.d. assumptions is violated. We show that, for nonparametric kernel based regression, classical model selection procedures such as cross-validation, generalized cross-validation and $v$-fold cross-validation break down in the presence of correlated data and not the chosen smoothing method. Therefore, we develop a model selection procedure for LS-SVM in order to effectively handle correlation in the data without requiring any prior knowledge about the correlation structure.

Next, we propose bias-corrected $100(1-\alpha)\%$ approximate confidence and prediction intervals (pointwise and uniform) for linear smoothers, in particularly for LS-SVM. We prove, under certain conditions, the asymptotic normality of LS-SVM. Further, we show the practical use of these interval estimates by means of toy examples for regression and classification.

Finally, we illustrate the capabilities of the proposed methods on a number of applications i.e. system identification, hypothesis testing and density estimation.

Promotor: Bart De Moor
Co-promotor: Johan Suykens

Wed 20 - Wed 20 Apr-11 SISTA Seminar - Alwin Stegeman, Laurent Sorber, Marco Signoretto
ESAT 00.62
11:15 am-1:00 pm
11:15-12:00 Alwin Stegeman (Univ. Groningen), "Three-way Decompositions - diverging components and how to avoid them"

12:00-12:30 Laurent Sorber (K.U. Leuven, Dept. Computer Science), "Optimization of real functions in complex variables"

12:30-13:00 Marco Signoretto (K.U. Leuven, ESAT-SCD), "Convex Multilinear Estimation and Non-parametric Tensor-based Models"

-Abstracts-

Alwin Stegeman, "Three-way Decompositions - diverging components and how to avoid them"

Fitting a three-way decomposition (CP: Canonical Polyadic, or
Candecomp/Parafac) with R components to a three-way array (or order-3
tensor) Z is equivalent to finding a best rank-R approximation of Z.
Contrary to the two-way case (the Singular Value Decomposition for
matrices), such a best rank-R approximation may not exist. This is
because the set of three-way arrays with rank at most R is not closed.
In this case, trying to compute a best rank-R approximation results in
diverging components. To avoid this problem, it has been proposed to
find a best approximation from the closure of the rank-R set instead.
For IxJx2 arrays and R<min(I,J) this can be done by using the
Generalized Schur Decomposition. For IxJxK arrays and R<min(I,J,K),
we propose the following method. Let the three-way decomposition (A,B,C)
feature diverging components. We show that (A,B,C) can be rewritten as a
decomposition in block terms, where each block term corresponds to a
group of diverging components. Moreover, we show that if the diverging
components occur in groups of two or three, then the limiting boundary
point X (which is a best approximation of Z from the closure of rank-R
set) can be obtained by fitting a block term decomposition to Z, in
which the core arrays of the blocks have a sparse canonical form. When
fitting this decomposition to Z, we use the block term decomposition of
(A,B,C) as initial value. We demonstrate our method by a simulation
study.

---

Laurent Sorber, "Optimization of real functions in complex variables"

Complex numbers are a fundamental tool for applied mathematics and many
engineering applications such as control theory, signal processing and
electrical engineering. Many nonlinear optimization methods use a first-
or second-order approximation of an objective function to generate a new
step or descent direction. A problem that arises in applying these
methods to real functions of complex variables, is that they are
necessarily nonanalytic in their argument, i.e. their Taylor series
expansion does not exist. A common workaround is to convert the
optimization problem to the real domain so that standard optimization
methods can be applied. We show that real functions in complex variables
do have a Taylor series expansion in complex variables, which we then
use to generalize existing optimization methods. We then apply these
methods to a number of case studies which show that complex Taylor
expansions give greater insight in the structure of the problem and that
this structure can often be exploited to improve computational
complexity and memory cost.

---

Marco Signoretto, "Convex Multilinear Estimation and Non-parametric
Tensor-based Models"

Tensor-based techniques are mostly based on decompositions that to
some extent generalize the matrix SVD. As such, the largest part of
the existing approaches relates to unsupervised methods. In this
presentation we discuss a different view inspired by machine learning
techniques. We examine a broad class of non-smooth convex optimization
problems for input patterns represented as tensors. A penalty based on
nuclear norms is used to enforce solutions with small (multilinear)
ranks. We show how an algorithm - termed Convex MultiLinear Estimation
(CMLE) - can be specialized to accomplish different data-driven
modeling tasks, both unsupervised and supervised. The arising models
are prune to successive analysis and interpretation; however they
might suffer from limited discriminative power. We then discuss
integration with kernel methods to overcome this limitation. We
introduce a novel family of structure-preserving product kernels for
tensors, illustrate its properties and show experimental results.

Wed 26 - Wed 26 Jan-11 Quarterly Meeting of OPTEC WG3
ESAT 00.62
9:00 am-11:00 am
"Quarterly Meeting of OPTEC Working Group 3 on Parameter and State Estimation"

Agenda:

1. Welcome (5 minutes)
2. Microbial dynamics under heat stress: liquid systems vs solid matrixes (Eirini Velliou 20 minutes)
3. Basics and application of coupled local minimizers (Johan Suykens en Saartje Arnout 30 minutes)
4. Coffee break (15 minutes)
5. Update and discussion on WG3’s actions in 2011 (20 minutes)
6. Semi-parametric approaches in Magnetic Resonance Spectroscopy (Diana Sima 20 minutes)
7. Any other business (5 minutes)

Fri 7 - Fri 7 Jan-11 K.U. Leuven seminars on Optimization in Engineering - WG2 Session: joint projects between departments
ESAT 00.62
10:00 am-12:00 pm
WG2 data-driven modelling
Session: joint projects between departments
Fr. Jan 7 2011, 10:00-12:00
ESAT 00.62

10:00-10:30 Title: Data clustering of human-demonstrated rigid body motions.
Presenters: Joris De Schutter (PMA), Carlos Alzate (ESAT), Tinne De Laet (PMA)

10:30-10:50 Title: Comparing Partial Least Squares with Nuclear Norm-Based Optimization for the prediction of batch-end quality.
Presenters: Geert Gins (CIT), Marco Signoretto (ESAT)

10:50-11:10 Title: Modelling of the environmental influence on a bridge's dynamic characteristics.
Presenters: Gersom Wursten (BWK), Tillmann Falck (ESAT)

11:10-11:30 Title: Identification of a Pilot Scale Distillation Column: A Kernel Based Approach.
Presenters: Bart Huyck (CIT), (Kris De Brabanter (ESAT))

11:30-11:50 Title: Block component analysis.
Presenters: Lieven De Lathauwer (ESAT), (Marc Van Barel (CS), Laurent Sorber (CS))

Wed 17 - Wed 17 Nov-10 Quarterly Meeting of OPTEC WG3
BOKU 04.20
9:00 am-11:00 am
"Quarterly Meeting of OPTEC Working Group 3 on Parameter and State Estimation"
The agenda is the following:
1. Welcome and agenda ( 5 min )
2. Intro of new members and update ( 20 min )
3. Basics and application of Moving Horizon Estimation (Jeroen Vandersteen 30 min)
4. Break ( 10 min )
5. Software used for parameter estimation in:
- civil engineering applications ( Geert Lombaert 15 min )
- bioprocesses ( Dominique Vercammen 15 min )
6. Lab visit to BOKU lab ( ... )

Wed 10 - Wed 10 Nov-10 OPTEC-MaNet Seminar on “Distributed network estimation and control problems” – Carlo Savorgnan
ESAT Aud.B
4:30 pm-6:00 pm
“Distributed multiple shooting for large-scale systems”

Speaker: Carlo Savorgnan, K.U.Leuven Optimization in Engineering Center (OPTEC)

Discussant: Bram Cornelis, ESAT-SCD (SISTA)

Manufacturing systems, process plants and networked systems are often composed by several subsystems whose dynamics are coupled through input-ouput connections. This talk introduces a generalization of the direct multiple shooting method which exploits the structure of the problem to achieve a highly parallelizable algorithm. The talk will provide an introduction to nonlinear model predictive control and direct multiple shooting.

Tue 9 - Tue 9 Nov-10 K.U. Leuven seminars on Optimization in Engineering (WG2) - Tillmann Falck, Jan Luts, Xinhai Liu, Alexander Caicedo Dorado, Kris De Brabanter
esat 91.91
2:00 pm-3:30 pm
WG2 data-driven modelling
Research progress talks
Tu. Nov 9 14:00-15:30
ESAT 91.91

14:00 Segmentation of nonlinear time series using convex optimization and kernels - Tillmann Falck (K.U. Leuven, ESAT-SCD)

14:15 An LS-SVM approach for classification of longitudinal data - Jan Luts (K.U. Leuven, ESAT-SCD)

14:30 Optimizing data fusion by tensor based hybrid clustering - Xinhai Liu (K.U. Leuven, ESAT-SCD)

14:45 Weighted LS-SVM for function estimation applied to artifact removal in biomedical signal processing - Alexander Caicedo Dorado (K.U. Leuven, ESAT-SCD)

15:00 Confidence Bands for Least Squares Support Vector Classification: A Regression Approach - Kris De Brabanter (K.U. Leuven, ESAT-SCD)

Thu 21 - Thu 21 Oct-10 K.U. Leuven Seminars on Optimization in Engineering - Alexander Bormann
ESAT 00.62
9:30 am-10:30 am

"EnerKite - Ways to cost-effective airborne wind energy converter"

Alexander Bormann

Abstract:

The topic of our talk will be "Ways to an cost-effective airborne wind turbine technology". The Presentation will show results of the preliminary EnerKite designs and reveal the importance of design principles, operational constraints and material selection.

Biographical Information
Alexander Bormann, CEO of aeroix, works in the field of wind engineering since 1993. He developed an optimized design method for the towers of multimegawatt turbines. Currently he and his company are developing technologies to make wind turbine towers useless. Aeroix is successful in kite automation since 2008. In addition to power generation Aeroix offers solutions for energy and material efficiency in aeronautics and architecture.

Thu 21 - Thu 21 Oct-10 2nd Meeting of OPTEC's Kite Power Project
ESAT 00.62
10:30 am-12:00 pm
2nd Meeting of OPTEC's Kite Power Project

Thu 14 - Thu 14 Oct-10 SISTA Seminar - Kim Batselier
ESAT Aud B
4:30 pm
"Maximum Likelihood Estimation and Polynomial System Solving"
Kim Batselier (K.U. Leuven, ESAT-SCD)

Discrete statistical models are probably one of the most important tools in bioinformatics. Learning the model parameters for these models from observations is commonly done via a maximum likelihood principle. In most cases however there are many solutions to this problem and only a local maximum can be found. The Expectation Maximization algorithm is the method of choice to tackle this problem. This talk presents a method that allows to find the global maximum likelihood estimate. The focus is limited on a specific class of discrete statistical models. For these models it is shown that the maximum likelihood estimates correspond with the roots of a multivariate polynomial system. Then, a new algorithm is presented, set in a linear algebra framework, which allows to find all these roots by solving a generalized eigenvalue problem. An illustrative example is worked out in which DNA is modeled in order to identify CpG islands.

Tue 28 - Tue 28 Sep-10 SISTA Seminar - Jorge Lopez
esat 00.57
2:00 pm
"Sparsifying LS-SVM models via L0 norm minimization"

Wed 8 - Wed 8 Sep-10 Doctoral Presentation - Toni Barjas Blanco
Auditorium of the Arenberg Castle
2:00 pm
"The river Demer controlled by MPC"
Toni Barjas Blanco (K.U. Leuven, ESAT-SCD)

Abstract
Flooding of rivers is a worldwide problem with severe consequences. This is also the case in Belgium where the Demer river causes floods in its basin during periods of heavy rainfall. In order to reduce these floods, the local water administration increased the storage capacity of the river with flood reservoirs to store the excessive amount of water during periods of heavy rainfall. Also hydraulic structures were added to control the discharges and water levels in the river, but also to control the flow of water from and into the reservoirs. Nowadays, these structures are controlled by an advanced three-position controller that determines the control actions based on logical rules that were derived in a heuristic way. Though these measures have significantly reduced the amount of flood damage in the Demer basin, simulations of historical rainfall events
on a full hydrodynamic model (InfoWorks) have shown that the flood damage could have been reduced even more if the hydraulic structures would have been controlled differently. Therefore, the goal of this thesis is to determine a more advanced control strategy that performs better than the currently used three-position controller.

In this thesis it is investigated whether a model predictive controller performs better with respect to flood regulation than the three-position controller. In order to do so, a simplified model is derived that is fast and accurate enough to be used in a model predictive control framework. The simplified model is of the reservoir type and is calibrated and validated based on simulation data obtained from simulating historical rainfall events with the InfoWorks model of the Demer river. Next, a nonlinear
model predictive controller is described that uses the simplified model in order to determine the future control actions. In contrast to standard model predictive control schemes for setpoint regulation we have chosen to use the gate levels as control inputs instead of the discharges over the gates because the nonlinear gate dynamics cannot be neglected during flood events. The proposed scheme also tackles problems like local uncontrollability of the gates and constraint infeasibility.

A nonlinear moving horizon estimator is also added to the model predictive control scheme. Based on historical measurements this estimator determines the most probable value of the actual state of the system. This estimate is then passed to the model predictive controller that uses this state information for its predictions. The control scheme resulting from
the combination of the model predictive controller with the moving horizon estimator is then compared with the three-position controller by comparing their respective performance in simulations based on historical rainfall data. Robustness of the new scheme is also tested by adding a realistic amount of uncertainty to the rainfall predictions.

The thesis ends with a theoretical contribution in the stability of model predictive control. More specific, a new algorithm is described to determine low-complexity polytopic invariant sets. The proposed algorithm is then used for improved setpoint regulation of the upstream part of the Demer river.

Promotors
Prof. dr. ir. B. De Moor, promotor
Prof. dr. ir. J. Berlamont, co-promotor

Tue 7 - Tue 7 Sep-10 SISTA Seminar - Philippe Dreesen, Kim Batselier, Xinhai Liu, Pieter Penninckx
ESAT 01.60
2:00 pm-3:30 pm
14:00 Solving Systems of Polynomial Equations
Philippe Dreesen, Kim Batselier, Bart De Moor

We discuss a problem coming from algebraic geometry, namely finding the solutions of a system of multivariate polynomial equations. Although the emphasis in this contribution is not directly on tensor algebra, it is well known that (homogeneous) polynomial equations are closely related to tensors: a polynomial of degree one is a linear form, a polynomial of degree two is a quadratic form, and in the same manner, a polynomial of degree k can easily be identified with a symmetric k-tensor.

The problem of solving a system of polynomial equations is ubiquitous in applied mathematics, science and engineering. Although there exists a huge body of literature in algebraic geometry, computational methods very often employ symbolic algebra methods to tackle this problem, which typically result in numerically poorly conditioned problems. Only a few decades ago it was (re-)discovered that solving a system of polynomial equations can be tackled using a (numerical) linear algebra framework, using eigenvalue computations. We will present a method which phrases the problem at hand as a problem in linear algebra and realization theory, employing only matrix computations.

The presented method proceeds in three steps. First, the problem is linearized by separating the coefficients from the monomials: we construct a Sylvester-like coefficient matrix which is multiplied with a vector containing monomials to obtain a system of linear equations. Next, the number of solutions is determined from the corank of this matrix, where one has to take care in separating the affine solutions from the solutions at infinity. Finally, the solutions themselves are determined from eigenvalue computations. This is done by taking into account the specific structure in the monomial basis which naturally leads to a method where realization theory can be applied in the kernel of the coefficient matrix to find the roots.

We will highlight applications in optimization theory and system identification. Future work and open problems will also be discussed.

14:30 Hybrid Clustering of Multi-view Data via MLSVD
Xinhai Liu, Lieven De Lathauwer, Wolfgang Glänzel, Bart De Moor

We present a hybrid clustering framework of multi-view data via tensor methods, which can be regarded as an extension of the spectral clustering based on modularity maximization. This hybrid clustering can be solved by both the truncated multilinear singular value decomposition (MLSVD) and higher-order orthogonal iteration (HOOI). So two algorithms are formed: average hybrid clustering by multilinear singular value decomposition (AHC-MLSVD) and weighted hybrid clustering by higher-order orthogonal iteration (WHC-HOOI). In particular, WHC-HOOI can leverage the effect of data from various views. Experimental results conducted on the synthetic data and on a large journal set retrieved from the Web of Science (WoS) database have demonstrated the effectiveness of our algorithms.

Pieter Penninckx, Lieven De Lathauwer

JADE is a tensor decomposition algorithm that manipulates a tensor along two of its modes. We show that in applications in independent component analysis, there is an extra structure in the tensor decomposition. We present an adapted version of the JADE algorithm that exploits this structure by additionally manipulating the tensor along its third mode. We conclude that this approach, although promising in theory, does not lead to better results in practice.

Thu 5 - Thu 5 Aug-10 OPTEC Working Group 1 Meeting (Dynamic and Embedded Optimization)
ESAT 00.57
2:00 pm-4:00 pm
Trimonthly Meeting of OPTEC Working Group 1 on Dynamic and Embedded Optimization
Aim of the meeting is to update the group members on each others activities. I will consist of a plenary part with an update, a short round of what each member currently works on, and a general talk on embedded optimization by M. Diehl. Then, after a break, we will focus on APPLICATIONS, and perform a smaller group brainstorming on what each member can contribute to these applications. Particular focus is on two of OPTEC's main application projects, kite power and robot machine interaction. Last, we prepare the next meeting within the kick-off of OPTEC-II.

Thu 15 - Thu 15 Jul-10 Simon Stevin Lecture on Optimization in Engineering - Mung Chiang
Auditorium of the Arenberg Castle
5:00 pm-6:00 pm
15th Simon Stevin Lecture on Optimization in Engineering

"Optimization in Networking"

Mung Chiang
Electrical Engineering Department, Princeton University

slides

Abstract

Optimization theory has provided both a modeling language and solution methodologies to a wide range of problems in communication networks. Recent successes include P2P streaming, TCP congestion control, IP routing, wireless scheduling, and power control. This talk surveys the current state and latest results on the applications of distributed, stochastic, robust, nonconvex, and combinatorial optimization in networking, and on the emergence of a first-principle based network design perspective enabled by the “optimization way of thinking”. Throughout the talk, we highlight both mathematical challenges raised by these applications and practical impact made by theory to the Internet and wireless networks.

Biographical Information

Mung Chiang is an Associate Professor of Electrical Engineering, and an Affiliated Faculty of Applied and Computational Mathematics and of Computer Science, at Princeton University. He received the B.S. (Honors) in Electrical Engineering and Mathematics, M.S. and Ph.D. degrees in Electrical Engineering from Stanford University in 1999, 2000, and 2003, respectively, and was an Assistant Professor at Princeton University 2003-2008. His research areas include optimization, distributed control, and stochastic analysis of communication networks, with applications to the Internet, wireless networks, broadband access networks, content distribution, and network economics. His awards include Presidential Early Career Award for Scientists and Engineers 2008 from the White House, TR35 Young Innovator Award 2007 from Technology Review, Young Investigator Award 2007 from ONR, Young Researcher Award Runner-up 2004-2007 from Mathematical Programming Society, CAREER Award 2005 from NSF, as well as Frontiers of Engineering Symposium participant 2008 from NAE and Engineering Teaching Commendation 2007 from Princeton University. He was a Princeton University Howard B. Wentz Junior Faculty and a Hertz Foundation Fellow. His paper awards include ISI citation Fast Breaking Paper in Computer Science, IEEE GLOBECOM Best Paper three times, and IEEE INFOCOM Best Paper finalist. His guest and associate editorial services include IEEE/ACM Trans. Netw., IEEE Trans. Inform. Theory, IEEE J. Sel. Area Comm., IEEE Trans. Comm., IEEE Trans. Wireless Comm., and J. Optimization and Engineering. He has filed 16 patents and co-chaired the 38th Conference on Information Sciences and Systems.

The "Simon Stevin Lecture Series on Optimization in Engineering" is set up in order to promote optimization in engineering. For this aim, every quarter of the year an outstanding international scholar is invited to report on latest progress in the development of optimization algorithms and their applications in engineering.
Simon Stevin (1548-1620) was a Flemish mathematician and engineer. Among other, he helped to advance the use of decimal fractions, was the first to explain the tides by the attraction of the moon, and discovered the hydrostatic paradox. He made numerous inventions, among them a wind propelled carriage with sails, the "land yacht", which once impressed Prince Maurice of Orange as it moved faster than horses, in around 1600 on the beach between Scheveningen and Petten. Simon Stevin was fond of promoting the use of science in daily life and in craftmanship, and translated various mathematical terms into dutch. Among other, he introduced the dutch word for mathematics, "wiskunde".

Directly after this Simon Stevin Lecture, a little reception will be given at 18:00 in the salons of Arenberg Castle, to which all attendants of the lecture are most warmly welcome!

Wed 14 - Wed 14 Jul-10 K.U. Leuven Seminars on Optimization in Engineering - Julio R. Banga
ESAT 01.57
9:00 am-10:00 am

"MIXED-INTEGER DYNAMIC OPTIMIZATION IN SYSTEMS BIOLOGY"

Julio R. Banga, IIM-CSIC, Vigo, Spain

Abstract

Mathematical optimization aims to make a system or design as effective or
functional as possible, computing the quality of the different
alternatives using a mathematical model. Most models in systems biology
have a dynamic nature, usually described by sets of differential
equations. Dynamic optimization addresses this class of systems, seeking
the computation of the optimal time-varying conditions (control variables)
to minimize or maximize a certain performance index. Dynamic optimization
can solve many important problems in systems biology, including optimal
control for obtaining a desired biological performance, the analysis of
network designs and computer aided design of biological units.

In this talk I will describe our research in continuous and mixed-integer
dynamic optimization (MIDO) problems in the framework of computational
systems biology. I will also present results for
several challenging optimization problems related with bioreactor
optimization, optimal drug infusion
to a patient and the minimization of intracellular oscillations.

Wed 7 - Fri 9 Jul-10 11th Computer Applications in Biotechnology (CAB 2010)
Leuven
9:00 am-5:00 pm

Mon 5 - Wed 7 Jul-10 9th International Symposium on Dynamics and Control of Process Systems (DYCOPS 2010)
Leuven
9:00 am-5:00 pm

Wed 23 - Thu 24 Jun-10 Interdisciplinary Privacy Course 2010
Computer room at the Mediacentre (Faculty of Social Sciences)
9:00 am-6:00 pm

This interdisciplinary course is part of the thematic training of the Leuven Arenberg Doctoral School Training Programme and supported by IAP BCRYPT and LICT. The course is mainly aimed at Ph.D. students from all disciplines (either from the K.U.Leuven or from other universities,) but also open to undergraduate students, post-docs, people working in industry, or anyone else interested on the topic.

The course will provide an overview of various aspects of privacy from the technical, legal, economics, and social science perspectives. While the broad focus of the course is on privacy in electronic services, this year’s edition of the course will have a special focus on social networks.

## When

·            Wednesday, June 23, from 09:30 to 17:30

·            Thursday, June 24, from 09:00 to 18:00

## Where

Computer room at the Mediacentre (Faculty of Social Sciences)

## Speakers

The course will last two days and consist of eight lectures. The lecturers include five speakers from different departments and faculties in K.U.Leuven and an invited speaker:

·            Prof. Alessandro Acquisti, (Carnegie Mellon University, USA)

·            Prof. Bettina Berendt, Computer Science (K.U.Leuven)

·            Dr. Claudia Diaz, Electrical Engineering (K.U.Leuven)

·            Dr. David Geerts, Faculty of Social Sciences (K.U.Leuven)

·            Seda Gürses, Electrical Engineering / Computer Science (K.U.Leuven)

·            Eleni Kosta, Faculty of Law (K.U.Leuven)

## Registration

·            The course is free of charge, but attendees are required to register by sending an email to claudia.diaz@esat.kuleuven.be

·            The registration deadline is: Tuesday, June 15

If you have any questions or would like to know more information please send an email to claudia.diaz@esat.kuleuven.be.

# Programme

## Wednesday, June 23

09:30 Introduction (Claudia Diaz)

10:30 Coffee break

11:00 Overview of Privacy Enhancing Technologies (PETs)  (Claudia Diaz)

12:30 Lunch break

14:00 Exploring European data protection: From social networks to cookies (Eleni Kosta)

15:30 Coffee break

16:00 Privacy and Web mining (Bettina Berendt)

17:30 End

## Thursday, June 24

09:00 To share or not to share - a user's perspective on privacy in social networks (David Geerts)

10:30 Coffee break

11:00 Privacy Concerns and Information Disclosure: An Illusion of Control Hypothesis (Alessandro Acquisti)

12:30 Lunch break

14:00 Privacy, Requirements Engineering and Online Social Network Services (Seda Gürses)

15:30 Coffee break

16:00 Predicting Social Security Numbers From Public Data (Alessandro Acquisti)

17:30 Discussion speakers and participants

18:00 end

[PDF]

Mon 14 - Mon 14 Jun-10 OPTEC Course on Numerical Methods for Optimal Control (Lecture 7) - Moritz Diehl
ESAT 01.57
4:00 pm-6:00 pm
OPTEC Course on Numerical Methods for Optimal Control (Lecture 7)
"Inexact SQP Methods / MPC stability"

General info: This 7 lecture course is aimed at all OPTEC members that are interested in developing numerical methods for optimal control. It introduces the attendants into both, well-known and brand-new algorithmic ideas that are important for the optimal control software that is developed at OPTEC. It will also introduce into the technical vocabulary in the field.

Some non-authorized notes related to the lecture by M. Diehl on inexact SQP methods can be found here.

Thu 10 - Thu 10 Jun-10 K.U. Leuven seminars on Optimization in Engineering (WG2) - Jonas Sjoberg
ESAT 01.60
2:00 pm
"Nonlinear System Identification, on Algorithms Finding Initial Parameter Values for the Iterative Minimization of the Cost Function"

Jonas Sjoberg (Chalmers University, Sweden)

Computing the parameter estimate in a Maximum Likelihood system identification problem typically leads to an iterative minimization of the negative log-likelihood function. Only for models which are linear in the parameters one can be sure of avoiding local minima. A good start value for the iterative minimization decreases the risk of being trapped in a local minimum. This talks starts with a tutorial part explaining the described problem in some detail. Examples are given on earlier work on algorithms for obtaining initial estimates for linear models such as ARMAX and state space models. We then turn to nonlinear system identification. Initialization of nonlinear black-box models using incremental models is explained and motivated. Algorithms for the traditional nonlinear structures Hammerstein and Wiener models are covered and a new algorithm for initializing Wiener-Hammerstein models is presented. This new algorithm is used on the benchmark data from SYSID 2009 and the result is analyzed and explained.

Host: WG2 - Johan Suykens, K.U. Leuven, ESAT-SCD

Tue 11 - Tue 11 May-10 SISTA Seminar - Ingrid DAUBECHIES
Thermotechnisch Instituut, Kasteelpark Arenberg 41, 3001 HEVERLEE, Aula Van de Tweede Hoofdwet
11:00 am

ANNOUNCEMENT of  a K.U.Leuven Seminar, Campus ARENBERG,

Prof.  Ingrid DAUBECHIES
(Princeton university, USA)

“Surfing with Wavelets”

Tuesday May 11, 2010 from 11h00 to 12h00

Thermotechnisch Instituut, Kasteelpark Arenberg 41, 3001 HEVERLEE
Aula Van de Tweede Hoofdwet (TI 01.0002)

ABSTRACT:

Wavelets are used in the analysis of sounds and images, as well as in many other applications. The wavelet transform provides a mathematical analog to a music score: just as the score tells a musician which notes to play when, the wavelet analysis of a sound takes things apart into elementary units with a well defined frequency (which note?)  and at a well defined time (when?). For images wavelets allow you to first describe the coarse features with a broad brush, and then later to fill in details. This is similar to zooming in with a camera: first you can see that the scene is one of shrubs in a garden, then you concentrate on one shrub and see that it bears berries, then, by zooming in on one branch, you find that this is a raspberry bush. Because wavelets allow you to do a similar thing in more mathematical terms, the wavelet transform is sometimes called a "mathematical microscope".

Wavelets are used by many scientists for many different applications.  Outside science as well, wavelets are finding their uses: wavelet transforms are an integral part of the image compression standard JPEG2000.

The talk will start by explaining the basic principles of wavelets,  which are very simple. Then they will be illustrated with some examples, including an explanation of image compression.

BRIEF BIOGRAPHY:

Ingrid Daubechies received both her bachelor's and Ph.D. degrees (in 1975 and 1980) from the Free University in Brussels, Belgium.  She held a research position at the Free University until 1987. From 1987 to 1994 she was a member of the technical staff at AT&T Bell Laboratories, during which time she took leaves to spend six months (in 1990) at the University of Michigan, and two years (1991-93) at Rutgers University. She is now at the mathematics department and the Program in Applied and Computational Mathematics at Princeton University.  She was awarded a Leroy P. Steele prize for exposition in 1994 for her book Ten Lectures on Wavelets. From 1992 to 1997 she was a fellow of the John D. and Catherine T. MacArthur Foundation. She is a member of  the American Academy of Arts and Sciences, the American Mathematical Society, the Mathematical Association of America, the Society for Industrial and Applied Mathematics, and the Institute of Electrical and Electronical Engineers. She is married and has two children.

ORGANIZERS:

The seminar is organized by the Departments of Electrical Engineering, Mathematics and Computer Science on the occasion of the appointment of Prof. Daubechies as International Francqui Professor at the Free University in Brussels (January-June 2010).

Mon 10 - Mon 10 May-10 Doctoral Presentation - Peter Karsmakers
Auditorium of the Arenberg Castle
2:00 pm
Sparse kernel-based models for speech recognition

Peter Karsmakers (K.U. Leuven, ESAT-SCD)

Abstract

Machine learning methods estimate models of the reality based on a set of examples. Kernel-based methods are a specific family in this context which obtain state-of-the-art performance for a wide range of application tasks such as in bioinformatics, financial engineering, or time series prediction. Those techniques are however not (yet) found often within the domain of speech recognition. Therefore this thesis studies the integration of a particular family of machine learning methods into an Automatic Speech Recognition (ASR) framework.

ASR typically involves the task of transforming a spoken speech utterance into a set of speech units (e.g. phonemes) which then are combined as words, sentences. The speech units can be linked to acoustical signals of variable length. Therefore, the boundaries of speech units are not easily identified in general.

The following 4 issues were taken into account when new kernel-based methods are designed: (i) given the fact that speech units can be spoken in many different ways, a large set of examples is needed to estimate good models which means that new learning methods need to cope with a large amount of data; (ii) as described above, variable length signals need to be processed; (iii) a probabilistic interpretation of the recognized speech units is more convenient for integration with other ASR units, such as a knowledge base containing rules of grammar; (iv) kernel-based methods are typically formulated for binary classification problems, however in ASR more classes (e.g. 39 phonemeclusters) need to be distinguish.

The manuscript elaborates in detail on 2 original methods which satisfy the above requirements. The first one focuses on the last two issues. The second method aims at an as compact solution as possible to obtain short evaluation times. For this purpose a generic method was designed to compute approximate sparse solutions to a set of linear equations. Experiments indicate that, when using these new methods in an ASR benchmark case, state-of-the-art results can be obtained.

Promotors
Prof. dr. ir. J.A.K. Suykens
Prof. dr. ir. H. Van hamme

Mon 26 - Wed 5 May-10 SOCN course on LMI Optimization with Applications in Control - Didier Henrion
Thermotechnisch Instituut - Aula van de Tweede Hoofdwet 01.02
9:00 am-12:00 pm
"LMI Optimization with Applications in Control"
by Didier Henrion (Laas CNRS-Toulouse, FRANCE)

This is a course for graduate students or researchers with a background in linear control systems, linear algebra and convex optimization.
The focus is on semidefinite programming (SDP), or optimization over linear matrix inequalities (LMIs), an extension of linear programming to the cone of positive semidefinite matrices. Since the 1990s, LMI methods have found numerous applications mostly in combinatorial optimization, systems control and signal processing.

The course will be presented over 6 lectures between April 26 and May 5, 2010. Lectures will take place in Leuven.

Dates : April 26 (afternoon), 27, 28 and May 03 (afternoon), 04, 05.
(afternoon: 2:00-5:00)

Thu 18 - Thu 18 Mar-10 SISTA Seminar - Laszlo Gyorfi
ESAT Aud A
4:00 pm-5:00 pm
"Nonparametric prediction of stationary time series"

Prof. Laszlo Gyorfi (Budapest University of Technology and Economics)

Abstract:
For the prediction of stationary time series, the standard nonparametric
techniques can be applied if the time series is weakly dependent, for
example, there is a kind of mixing. In the general case, universally
consistent prediction can be constructed via machine learning type
aggregation of several nonparametric regression estimates. For such
aggregation, we show the exponential weighting.

Application areas: nonparametric statistics, system identification, machine learning and data mining

Short CV: László Györf. is professor at the Department of Computer Sciences and Information Theory at Budapest University of Technology and Economics and is the head of the Research Group for Informatics and Electronics of the Hungarian Academy of Sciences.
He obtained the diploma in Mathematics and Physics and two PhDs from the University of Budapest in 1970, 1974 and 1988. Since 1997 he is fellow of IEEE and since 2001 he is also Ordinary Member of the Hungarian Academy of Sciences. He is also the winner of numerous prizes such as Farkas Gyula Prize, Pollák - Virág Prize, Jacob Wolfowitz Prize and Széchenyi Prize. His main interests are stochastic approximation, pattern classification, nonparametric density, regression and entropy estimation, mathematical statistics, prediction of time series, multiple access communication, source coding and empirical portfolio selection.
He is the (co)-author of more than 100 papers in various fields of statistics and of several books such as Nonparametric density estimation: The L1 view, Nonparametric Curve Estimation from Time Series, A Probabilistic Theory of Pattern Recognition, A Distribution-Free Theory of Nonparametric Regression, Principles of Nonparametric Learning (ed.) and Multiple Access Channels: Theory and Practice (ed.).

[PDF]

Sat 13 - Sat 20 Mar-10 Course on Embedded and Convex Optimization for Control - Stephen Boyd and Moritz Diehl
--
9:00 am-4:00 pm
"Course on 'Embedded and Convex Optimization for Control'
by Stephen Boyd and Moritz Diehl,
K.U. Leuven, March 15-19, 2010.
Details on the course can be found on the following webpage:
Embedded and Convex Optimization for Control

Wed 10 - Wed 10 Mar-10 OPTEC Course on Numerical Methods for Optimal Control (Lecture 1) - Moritz Diehl
Esat Aud. B
10:00 am-12:00 pm
This lecture is aimed to all OPTEC members that are interested in developing numerical methods for optimal control. It introduces the attendants into both, well-known and brand-new algorithmic ideas that are important for the optimal control software that is developed at OPTEC. It will also introduce into the technical vocabulary in the field.

Topics to be covered in the course:
- optimal control problem formulations
- direct methods, sequential and simultaneous
- collocation, direct single and multiple shooting
- inexact SQP and SCP methods for optimal control
- QP linear algebra: condensing, sparse factorizations
- parametric optimization and real-time iterations

The first lecture will be followed by more detailed ones  on a biweekly basis.

slides(Numerical Optimal Control)
slides(Dynamic Programming)
slides(Direct Methods)

Tue 2 - Tue 2 Mar-10 Simon Stevin Lecture on Optimization in Engineering - Matthias Heinkenschloss
Auditorium of the Arenberg Castle
5:00 pm-6:00 pm
14th Simon Stevin Lecture on Optimization in Engineering

"Numerical Solution of PDE Constrained Optimization Problems"

Matthias Heinkenschloss (Rice University)
www.caam.rice.edu/~heinken/

Flyer, poster
Abstract

Optimization problems governed by partial differential equations (PDEs) arise in many science and engineering applications in the form of parameter identification problems, optimal design problems, or optimal control problems. Ultimately the numerical solution of these problems lead to large scale nonlinear programming problems (NLPs) in which the objective function or some of the constraint functions depend on the discretized PDE. However, for the reliable and efficient solution of PDE constrained optimization problems it is not enough to couple PDE discretizations with large scale NLP solvers. Instead, it is important to carefully integrate the analysis of the infinite dimensional optimization problems, the PDE discretization, and the optimization algorithms. In this talk I will present some PDE constrained optimization applications and discuss their numerical solution. I will emphasize the differences between the numerical solution of a single PDE and the PDE constrained optimization problem.

Biographical Information

Matthias Heinkenschloss is Professor at the Department of Computational and Applied Mathematics at Rice University. He obtained the diploma and PhD from the University of Trier. He currently is Associate Editor of 'Systems & Control Letters', Associate Editor of 'Mathematical Programming, Series A' and Member of the Editorial Board of 'Numerical Linear Algebra with Applications'.
The research interests of Dr. Heinkenschloss are in optimization, optimal control and parameter identification, partial differential equations, model reduction and the application of optimization in science and engineering.

The "Simon Stevin Lecture Series on Optimization in Engineering" is set up in order to promote optimization in engineering. For this aim, every quarter of the year an outstanding international scholar is invited to report on latest progress in the development of optimization algorithms and their applications in engineering.
Simon Stevin (1548-1620) was a Flemish mathematician and engineer. Among other, he helped to advance the use of decimal fractions, was the first to explain the tides by the attraction of the moon, and discovered the hydrostatic paradox. He made numerous inventions, among them a wind propelled carriage with sails, the "land yacht", which once impressed Prince Maurice of Orange as it moved faster than horses, in around 1600 on the beach between Scheveningen and Petten. Simon Stevin was fond of promoting the use of science in daily life and in craftmanship, and translated various mathematical terms into dutch. Among other, he introduced the dutch word for mathematics, "wiskunde".

Directly after this Simon Stevin Lecture, a little reception will be given at 18:00 in the salons of Arenberg Castle, to which all attendants of the lecture are most warmly welcome!

***** REGISTRATION ENCOURAGED *****
Please send an e-mail with the subject "STEVIN" to optec.secretariaat@esat.kuleuven.be if you intend to participate in the event. No obligation, just to help us getting an idea how many people plan to come.

This Stevin lecture is co-sponsored by ICCoS (Identification and Control of Complex Systems), a Scientific Research Network of the Research Foundation - Flanders (FWO-Vlaanderen).

Fri 26 - Fri 26 Feb-10 K.U. Leuven Seminars on Optimization in Engineering - Saartje Arnout
BOKU 03.10
4:00 pm-5:30 pm
"A methodology for the optimal design of shell shaped structures" (WG4)
Saartje Arnout

In modern architecture, there is a tendency towards complex structures. These structures frequently have the shape of a classical shell. As the geometry of a shell plays a key role in its structural behaviour, this entails an increasing attention for the optimal design of shell structures.
Structural optimization is a very promising design tool. However, some crucial choices concerning the method, the design variables, the parametrization etc. have to be made to use it in practice, Therefore, a methodology is presented where optimization is applied throughout the design process to obtain an optimal final design. It is demonstrated how to use the available information in such a way that the results of the optimization are helpful in the current stage of the design process.
In the conceptual design stage, an initial design concept for the shape of the structure is usually available. However, no details about the structure are available. As a consequence, a simplified analysis model with only a small number of relevant load cases can be sufficient. This enables the use of global optimization algorithms. In general, the global optimum is targeted, corresponding to the least material use. Alternatively, the design space can be explored to see the variety of good or near-optimal solutions in the design space.
In later design stages, the overall geometry is fixed and large design changes have to be avoided. Consequently, local optimization algorithms maybe preferred. More details about the structure are known. Therefore, it is possible to implement the design criteria more strictly. In this stage, it can be necessary to account for uncertainties, such as initial geometrical imperfections of the shell structure. An optimization based on this detailed model with a limited design space can show the most efficient changes for reduction of the material volume.

Thu 28 - Thu 28 Jan-10 K.U. Leuven Seminars on Optimization in Engineering - Olivier Brüls
Esat 00.57
11:00 am-12:00 pm
"Recent developments in simulation, optimization and control of flexible multibody systems"
Olivier Brüls, University of Liège, Belgium

This talk addresses some recent extensions of the finite element approach for the analysis, design and control of flexible mechanisms. A particular attention is devoted to modular simulation concepts, advanced time integration methods, efficient sensitivity analysis and topology optimization problems. Practical applications of those techniques can be found in various fields of engineering, e.g. in automotive engineering (vehicle suspensions, powertrains), aerospace engineering (landing gears, flaps, deployable space structures), robotics, machine tools, biomechanics or biomedical instruments.

Firstly, an integrated simulation approach will be presented for articulated systems composed of rigid bodies, flexible bodies, kinematic joints, actuators, sensors and control units. I will focus on some numerical aspects concerning the time integration of the equations of motion which have the structure of strongly coupled differential-algebraic equations on a Lie group. The treatment of large rotation variables and the coupling between control state variables and mechanical generalized coordinates will be discussed in more detail.

Secondly, based on this simulation tool, a particular class of optimization problems in multibody dynamics will be considered, i.e. the topology optimization of structural components. Generally, topology optimization techniques use simplified quasi-static load cases to mimic the complex dynamic loadings in service. In contrast, I will present an optimization procedure which properly accounts for the actual dynamic interactions which occur during the motion of the flexible multibody system. The method relies on an efficient sensitivity analysis based on a semi-analytical direct differentiation approach. In order to illustrate the benefits of the proposed design approach, the optimization of a two degrees-of-freedom robot arm with flexible truss linkages will be analyzed. Finally, I will discuss some perspectives for the integrated control-structure optimization of multibody systems.

slides

Fri 22 - Fri 22 Jan-10 Exploitation of Research / Technology & Knowledge Transfer (Day 2)
(place to be confirmed)
This module introduces PhD students into the different routes of technology & knowledge transfer and highlights the key attention points and success factors. The three main aspects of research exploitation will be covered by experts in each field: 1) contract & collaborative research; 2) patenting & licensing; and 3) creating a new company (spin-off). The approach is highly hands-on. In addition to case studies & testimonies, small groups (2 to 4 people) of Ph.D. students will be coached to work out an exploitation plan for the research results of one of them / one of their research groups.

Day 2 - 22 January: place to be confirmed
09h00 - 11h00 Managing Intellectual Property Rights: General Framework
Prof. Marie-Christine Janssens, Centre for Intellectual Property Rights
11h00 - 12h00 Patenting & Licensing Strategies in a University Context
Dr. Ivo Roelants, IPR Officer K.U.Leuven R&D
12h00 - 13h00 Sandwich Lunch
13h00 - 14h00 How to Use Patent Databases: Introduction
Dr. Ivo Roelants, IPR Officer K.U.Leuven R&D
14h00 - 16h00 How to Use Patent Databases: Hands on Session
Dr. Ivo Roelants, IPR Officer K.U.Leuven R&D & collegues

[PDF]

Tue 12 - Tue 12 Jan-10 K.U. Leuven Seminars on Optimization in Engineering - Andrew Wagner
ESAT 91.91
3:00 pm-4:00 pm
"Towards a Practical Face Recognition System:  Robust Registration and Illumination by Sparse Representation"
Andrew Wagner, University of Illinois

Our ability to recognize people by their faces is crucial for our day-to-day interactions.  Automating this task would have many applications in security, entertainment, and human-computer interaction.  Unfortunately, while most contemporary face recognition algorithms work well under laboratory conditions, they degrade when tested in less controlled environments.  This is mostly due to the difficulty of simultaneously handling variations in illumination, alignment, pose, and occlusion. In this talk, I will propose a simple and practical face recognition system that achieves a high degree of robustness and stability to all these variations. Using tools from sparse representation and a projector-based training image acquisition system, it is possible to align a test face image with a set of frontal training images in the presence of signiﬁcant registration error and occlusion.

Slides

Wed 9 - Wed 9 Dec-09 Exploitation of Research / Technology & Knowledge Transfer (Day 1)
Thermotechnisch Instituut, Kasteelpark Arenberg 41, 3001 Heverlee
This module introduces PhD students into the different routes of technology & knowledge transfer and highlights the key attention points and success factors. The three main aspects of research exploitation will be covered by experts in each field: 1) contract & collaborative research; 2) patenting & licensing; and 3) creating a new company (spin-off). The approach is highly hands-on. In addition to case studies & testimonies, small groups (2 to 4 people) of Ph.D. students will be coached to work out an exploitation plan for the research results of one of them / one of their research groups.

Day 1 – 9 December, Thermotechnisch Instituut, Kasteelpark Arenberg 41, 3001 Heverlee
09h00 – 10h00: Introduction to Basic Tech Transfer Routes & Organisational Framework
Paul Van Dun, General Manager K.U.Leuven R&D,
Prof. Bart De Moor, Chairman IOF
10h00 – 11h30: The Innovation Process & Flemish Innovation Landscape
Prof. Koenraad Debackere, Managing Director K.U.Leuven R&D
11h30 – 12h15: Testimony: Added Value & Challenges University-Industry Interaction
Prof. Jan Delcour, Head of Laboratory on Food Chemistry & Biochemistry
12h15 – 13h15: Sandwich Lunch
13h15 – 14h45: Technology Market Assessment
Prof. Bart Van Looy, International Centre for Research on Entrepreneurship, Technology & Innovation Management
14h45 – 15h45: Technology Marketing
Ir. Wim Bens, Head of TU/e Innovation Lab
15h45 – 16h00: Closing remarksDr. Ir. Rudi Cuyvers, Innovation Manager K.U.Leuven R&D
16h00 – 18h00: Networking Drink

[PDF]

Tue 8 - Tue 8 Dec-09 K.U. Leuven Seminars on Optimization in Engineering - Joeben Bevirt
ESAT 00.57
11:00 am-12:00 pm
"Airborne wind turbines to harness high-altitude wind power"
Joeben Bevirt (Joby Energy, Santa Cruz, California)

Joby Energy is developing airborne wind turbines to harness high-altitude winds. Our technology will offer a solution to meet the growing, global demand for clean and low-cost energy. Operating at altitudes where winds
are considerably more consistent and powerful than surface-based winds, our system will produce twice the energy for the same rated power as surface-based turbines.
And compared to a surface-based wind turbine with the same rated power, our capital costs are considerably lower as our systems require approximately 1/40 the materials to build.
Lastly, our system will achieve a net capacity factor of approximately 70% resulting in a levelized cost of energy (LCOE) competitive with coal.

Tue 1 - Tue 1 Dec-09 K.U. Leuven Seminars on Optimization in Engineering - Martin Vetr
ESAT 00.57
5:00 pm-6:00 pm
"Waterbrakes for Engine Test-Benches "
Martin Vetr, JKU Linz (Austria)

Abstract:
Modern engine test-benches are a very expensive tool, yet they are necessary in the engine development. In many cases hydraulic dynamometers can be used as an alternative to more expensive electric motors.
In today's presentation a short introduction will be given to what a hydraulic dynamometer is and where it can be used. A first principles model will also be presented.

Wed 25 - Wed 25 Nov-09 Simon Stevin Lecture on Optimization in Engineering - Dominique Bonvin
Auditorium of the Arenberg Castle
5:00 pm-6:00 pm
13th Simon Stevin Lecture on Optimization in Engineering

"Adaptive Optimization in the Presence of Uncertainty"

Dominique Bonvin (EPFL, Lausanne)

Flyer, Poster

Abstract

This presentation discusses real-time optimization (RTO) strategies for improving the performance in the presence of uncertainty in the form of plant-model mismatch, drifts and disturbances. RTO typically uses a plant model to compute optimal inputs. In the presence of uncertainty, selected model parameters are estimated and the updated model is used for optimization. This two-step approach is repeated on-line as necessary. Although very intuitive, this approach suffers from the fact that the model is almost invariably inadequate, which prevents from reaching the plant optimum. Other approaches have been developed in the last two decades to overcome this difficulty. Recently, a generic formalization of these ad hoc fixes has been proposed under the label modifier adaptation. The basic idea is to leave the model parameters unchanged but to use the plant measurements to “appropriately” modify the optimization problem. The modifier adaptation approach will be presented and compared to the two-step approach, in particular with regard to the model adequacy condition. We will then go beyond this comparison and discuss different ways of using plant measurements for process improvement in the presence of uncertainty.
There are many questions to be addressed: (i) what can be done off-line prior to process operation, and what should be performed in real time, (ii) how much of the optimization effort is model-based and how much is data-driven, (iii) what to measure, what to adapt, how to adapt? We will then see that there exists another class of measurement-based optimization approaches that turns the optimization problem into a control problem; this class of methods includes NCO tracking, extremum-seeking control and self-optimizing control. A case study will illustrate the applicability of the various approaches.

Biographical Information:

Dominique Bonvin is Professor of Automatic Control and Dean of Bachelor and Master Studies at the Swiss Federal Institute of Technology in Lausanne (EPFL), Switzerland. He received his Diploma in Chemical Engineering from the ETH, Zürich, and his Ph.D. degree from the University of California, Santa Barbara. He worked in the field of process control for the Sandoz Corporation in Basel and with the Systems Engineering Group of ETH Zürich. He joined the EPFL in 1989, where his current research interests include modeling, identification and optimization of dynamic systems as well as process chemometrics. He served as Director of the Laboratoire d'Automatique for the periods 1993-97 and 2003-2007 and as Head of the Mechanical Engineering Department in 1995-97.

The "Simon Stevin Lecture Series on Optimization in Engineering" is set up in order to promote optimization in engineering. For this aim, every quarter of the year an outstanding international scholar is invited to report on latest progress in the development of optimization algorithms and their applications in engineering.
Simon Stevin (1548-1620) was a Flemish mathematician and engineer. Among other, he helped to advance the use of decimal fractions, was the first to explain the tides by the attraction of the moon, and discovered the hydrostatic paradox. He made numerous inventions, among them a wind propelled carriage with sails, the "land yacht", which once impressed Prince Maurice of Orange as it moved faster than horses, in around 1600 on the beach between Scheveningen and Petten. Simon Stevin was fond of promoting the use of science in daily life and in craftmanship, and translated various mathematical terms into dutch. Among other, he introduced the dutch word for mathematics, "wiskunde".

Directly after this winter's Simon Stevin Lecture, a little reception will be given at 18:00 in the salons of Arenberg Castle, to which all attendants of the lecture are most warmly welcome!

***** REGISTRATION ENCOURAGED *****
Please send an e-mail with the subject "STEVIN" to optec.secretariaat@esat.kuleuven.be if you intend to participate in the event. No obligation, just to help us getting an idea how many people plan to come.

This Stevin lecture is co-sponsored by ICCoS (Identification and Control of Complex Systems), a Scientific Research Network of the Research Foundation - Flanders (FWO-Vlaanderen).

Mon 16 - Fri 20 Nov-09 ATHENS Course "Numerical Optimal Control Algorithms, and Applications in Renewable Energy Systems"
Celestijnenlaan 200C, PC Room 00.04
9:00 am-6:00 pm
ATHENS Course "Numerical Optimal Control Algorithms and Applications in Renewable Energy Systems"
Organized by: Moritz Diehl, Boris Houska, and Joel Andersson

Thu 12 - Thu 12 Nov-09 SISTA Seminar - Philippe Dreesen
ESAT 01.57
4:00 pm
"From Solving Systems of Polynomial Equations to Linear Algebra"
Philippe Dreesen (K.U. Leuven, ESAT-SCD)

The task of solving systems of polynomial equations in several unknowns is one of the most classic problems of mathematics. Finding the solutions of systems of polynomial equations pervades science and engineering, and arises in system identification, control theory, optimization, signal processing, statistics, machine learning, mathematical biology and numerous other fields. It turns out that, behind the scenes, linear algebra and realization theory are key concepts in understanding and solving this problem. In this talk, we present a method in which the number of solutions of zero-dimensional polynomial systems corresponds to the corank of a certain affinely structured matrix built from the coefficients of the system, while all solutions can be obtained from the eigenvalues of a certain realization problem.

Tue 10 - Tue 10 Nov-09 Doctoral Presentation - Oscar Mauricio Agudelo
Auditorium of the Arenberg Castle
10:00 am
"The Application of Proper Orthogonal Decomposition to the Control of Tubular Reactors"
Oscar Mauricio Agudelo (K.U. Leuven, ESAT-SCD)

Abstract:

This dissertation considers two main research topics. First, this thesis explores the applicability of Proper Orthogonal Decomposition (POD) and Galerkin projection in the design of model predictive control schemes for tubular chemical reactors. These processes pose very interesting control problems, since their behavior is modeled by highly nonlinear Partial Differential Equations (PDEs), and they require the satisfaction of both their input and state constraints. In this study, POD is used together with Garlerkin projection for reducing the high-dimensionality of the discretized systems used to approximate the PDEs that model the reactors. Then, based on the resulting reduced-order models, Kalman filters and predictive controllers are designed. Although a significant model order reduction can be obtained with POD and Galerkin projection, these techniques do not reduce the number of state constraints (linear inequality constraints) which is typically very large. In this thesis we propose two methods to tackle this problem. In the first method we use univariate polynomials to approximate part of the basis vectors derived with the POD technique, and then we apply the theory of positive polynomials to find good approximations of the state constraints by linear matrix inequalities. In the second method, we exploit the similarities between the coefficients of consecutive state constraints for developing a greedy algorithm that selects a small number of constraints from the complete set.

The second main research subject of this thesis is related to speeding up the evaluation of reduced-order models derived by POD from nonlinear high-dimensional systems. Unlike the linear time-invariant case, the model-order reduction by POD and Galerkin projection does not conduce to an important computational saving when the high-dimensional models under consideration are nonlinear. Therefore, this thesis introduces two methods for coping with this situation. The first method takes advantage of the input-output nonlinear mapping capabilities, and the fast on-line evaluation of multi-layer perceptrons for accelerating the evaluation of the POD models. The second method exploits the polynomial nature of POD models derived from input-affine high-dimensional systems with polynomial nonlinearities, in order to generate compact and efficient formulations that can be evaluated much faster. Moreover, in this study it is shown how the use of sequential feature selection algorithms can provide a significant boost in the computational saving. Although this method is not as general as the first one, it might be applied to models with non-polynomial nonlinearities, provided that the nonlinearities can be approximated by low degree polynomials. In addition, conditions for guaranteeing the local stability of these POD models with polynomial nonlinearities are discussed.

Promotors: Prof. B. De Moor, Prof. J. Espinosa (UNAL), Prof. J. Vandewalle

Thu 22 - Thu 22 Oct-09 SISTA Seminar - Peter Karsmakers
ESAT 00.62
4:00 pm
"Large-Scale Kernel Logistic Regression for Speech Recognition"
Peter Karsmakers (K.U. Leuven, ESAT-SCD)

In this seminar we describe an efficient implementation of multi-class Kernel Logistic Regression (KLR) suitable for large-scale data sets which are common in speech recognition. Unlike the popular Support Vector Machines (SVM), KLR yields a-posteriori probabilities of membership for each of the classes and has a natural extension to the multi-class case. Compared to SVM, our implementation of KLR generates a much sparser model while obtaining comparable classification accuracies.  We motivate and validate the model for its use in a phoneme recognition system.  During speech recognition also incorrect elements which have no valid interpretation as lexical units (i.e., too long, too short, etc.), in SVM literature called universum data, are possible. We indicate that the use of unlabeled data in training the KLR model improves the final recognition performance.

Thu 15 - Thu 15 Oct-09 K.U. Leuven Seminars on Optimization in Engineering - Richard Braatz
ESAT 01.57
4:00 pm-5:00 pm
"Robust Optimal Control of Finite-time Distributed Parameter Systems "
Prof. Richard D. Braatz, Millennium Chair
University of Illinois at Urbana-Champaign, US,
braatz@uiuc.edu, http://brahms.scs.uiuc.edu

abstract:

Most products of high value such as in the pharmaceuticals, microelectronic, and nanotechnology industries are manufactured in a series of processing steps that operate over finite time. These processes are usually distributed parameter systems in which tight control is required. Computationally efficient methods are proposed for the robust optimal control of finite-time distributed parameter systems (DPS), in which robustness is ensured for either deterministic or stochastic parametric uncertainties. In the deterministic case, the effects of uncertainties on the states and product quality are quantified by power series expansions combined with linear matrix inequality or structured singular value analysis. In the stochastic case, the effects of uncertainties are quantified by power series or polynomial chaos expansions followed by Monte Carlo simulation. The robust performance analysis have been incorporated into fixed or model predictive control algorithms. The approaches are illustrated for several applications problems.

short bio:
Richard D. Braatz is Professor and Millennium Chair at the University of Illinois at Urbana-Champaign where he does research in the modeling, design, and control of chemical, pharmaceutical, and biomedical systems. He received MS and PhD degrees from the California Institute of Technology. Richard has consulted and/or collaborated with 15 companies including IBM, UTC Power, Eli Lilly, and Abbott Laboratories. Honors include the AACC Donald P. Eckman Award, ASEE Curtis W. McGraw Research Award, IEEE TCST Outstanding Paper Award, and Antonio Ruberti Young Researcher Prize. He is a Fellow of the Institute of Electrical and Electronics Engineers, the International Federation of Automatic Control, and the American Association for the Advancement of Science.

Mon 12 - Mon 12 Oct-09 K.U. Leuven Seminars on Optimization in Engineering - Matus Kopacka, Carlo Romani
ESAT 00.62
5:00 pm-6:30 pm
"Application of a MPC on a spark ignition engine"
Matus Kopacka

Abstract

In the following presentation I will be briefly introduced goals of my PhD study, particularly application of MPC on some control loops in the spark ignition engine control and also information about my person and home university.

"Hierarchical and distributed control structures for large scale systems"
Carlo Romani

Abstract

Nowadays there is a growing interest in the research on control of large scale systems as industrial plants, production and distribution networks, manufacturing systems. In particular an intensive stream of research concerns the development of distributed and multi-level control structures. These large scale systems can be difficult to control with a centralized control structure due to the computational complexity, to robustness and reliability problems and to communication bandwidth limitations. For these reasons many distributed control structures have been developed and applied. My research is focused on the following aspects:
• study of partitioning methods for distributed and hierarchical control. Decomposition of a dynamical system in a number of weakly interacting sub-systems and representation of a dynamical system at different levels of abstraction for the design of decentralized/distributed or hierarchical control systems;
• design of distributed and/or hierarchical control structures for large scale systems, often composed by many interacting subsystems. Focusing on the design of model predictive control (MPC).

We study the design of two layers hierarchical control system with Model Predictive Control. We consider a system described by a discrete-time, multi-input, multi-output Wiener model, that can be divided in two layers, the system to be controlled, S, and the actuators, Sact , characterized by two different time scales. We also give a mathematical formulation of the control design problem, in the case of three levels, and we present a possible solution with MPC that provide a unifying framework for different cases.

Wed 7 - Wed 7 Oct-09 K.U. Leuven Seminars on Optimization in Engineering - Kaisa Miettinen, Jussi Hakanen
CIT - PW00.01
11:00 am-12:00 pm
"Introduction to Some Interactive Methods for Multiobjective Optimization "
Kaisa Miettinen,
University of Jyvaskyla, Department of Mathematical Information Technology
kaisa.miettinen@jyu.fi

Many methods have been developed for multiobjective optimization. Typically, they aim at supporting a decision maker in finding the best compromise solution where several conflicting criteria are optimized at the same time. Because the compromise solutions, the so-called Pareto optima, cannot be ordered without additional information, the solution process requires preference information from a decision maker in some for or another, and methods can be classified in four classes based on the role of the decision maker. In this talk, we briefly characterize the different classes and pay most attention to interactive methods. In interactive methods, a solution patterns is formed and repeated several times, and at each iteration further information about the decision maker's preferences is inquired. In this way, the decision maker can learn about the nature of the problem and about the interdependencies among the criteria involved. (S)he can also adjust one's preferences while learning and concentrate on such solutions that seem most promising. We introduce two new interactive methods for solving nonlinear multiobjective optimization problems. The methods have been directed for different types of problems: Pareto Navigator for computationally expensive problems and Nautilus to enable free search (and also for group decision making). Finally, we briefly describe NIMBUS which has been applied in a variety of problems during the years.

"Some Applications of Interactive Multiobjective Optimization Methods"
Jussi Hakanen,
University of Jyvaskyla, Department of Mathematical Information Technology
jussi.hakanen@jyu.fi

In this talk, we describe some applications of interactive multiobjective optimization. The case studies arise from chemical engineering problems and the idea is to demonstrate how the field of chemical engineering can benefit from the tools of multiobjective optimization. By considering multiple objectives, the problem considered can be seen from different perspectives and there is no need for unnecessary simplifications just to make optimization possible. We introduce some real-world multiobjective optimization problems and describe how interactive multiobjective optimization methods have been utilized and how the methods have provided useful insight in the solution processes.

Bio's:

Jussi Hakanen got his PhD from the Department of Mathematical Information Technology, University of Jyväskylä, Finland in 2006. His research interests include theory, methods and applications of multiobjective optimization. During and after his doctoral studies, he has been working with applications related to chemical engineering including, e.g., paper making and wastewater treatment.

Kaisa Miettinen is a professor of industrial optimization at the Department of Mathematical Information Technology, University of Jyvaskyla, Finland, where she heads the research group on industrial optimization. Her research interests include multiobjective optimization (theory, methods and software), multiple criteria decision making, nonlinear programming, evolutionary algorithms, hybrid approaches as well as industrial applications. She is the President-Elect of the International Society on Multiple Criteria Decision Making.

Prof Kaisa Miettinen and Dr. Jussi Hakanen (Universiteit Jyväskyä, Finland) are visiting OPTEC October 6-8, 2009.

Wed 30 - Wed 16 Dec-09 Life Sciences Tech Watch doctoral course
MOLE Auditorium Oude Molen (MOLE 00.07), Kasteelpark Arenberg 50, 3001 Heverlee
3:30 pm-5:30 pm
Practical details:

- Timing: 1st semester academic year 09-10, starting September 30th, 11 courses of 2hrs.
Wednesday afternoons ; Time: 15:30 – 17:30

- Venue: (438-02) MOLE Auditorium Oude Molen, Kasteelpark Arenberg 50 , 3001 HEVERLEE

- Target public: Ph.D. students and postdocs of Group W&T, Group BMW and VIB.

- Evaluation: 9 out of 11 courses should be formally attended, proof of attendance

- Full program, speakers, location and registration to be announced on the BioSCENTer website and
on K.U.Leuven Agenda!

Website

Wed 30 - Wed 21 Oct-09 SOCN course on Model Predictive Control - Jan Maciejowski
Auditorium of the Arenberg Castle
9:15 am-12:15 pm
"Model predictive Control"
by Jan Maciejowski (University of Cambridge, UK)

Contents:

The course will start with the basic ideas of MPC, together with some specific examples of its advantages over “classical” control. It will then discuss the structure of MPC controllers, present possible variations (such as non-quadratic cost functions and stabilised predictions), and deal with important practicalities, especially disturbance feedforward and disturbance modelling. A state-space framework will be used, but the connection with the well-known GPC framework will be made. The course will then survey the state of recent MPC-related research, covering efficient computation, stability and robustness, prioritisation of objectives, the use of nonlinear models, the application of MPC to hybrid systems (which contain logic or mode switches as well as continuous dynamics), and distributed MPC. The course will be illustrated throughout with examples from various applications.

The course will be presented over 6 lectures between September 30 and October 21, 2009. Lectures will take place in Leuven.

Dates : September 30 and October 01, 02, 19, 20, 21.

Course slides

Wed 23 - Wed 23 Sep-09 SISTA Seminar - Gonzalo Acuna
ESAT Aud B
2:00 pm
"Towards appropriate training of Neural Networks and Support Vector Machines for complex dynamic processes"
Gonzalo Acuna (Universidad de Santiago de Chile, Comp. Eng. Dept.)

Abstract: With the increasing complexities of industrial and biological processes it becomes very difficult to build adequate first principle models in order to perform estimation, optimization or control among other important tasks. An alternative and fruitful approach to tackle this problem consists in designing appropriate data-driven models.  Neural Networks (NN) and  Support Vector Machines (SVM) have shown their usefulness as regression models. Their contribution to data-driven modelling will be highlighted by means of two examples: NN NARX modelling of  semiautogeneous mills in copper mining and SVM NARX modellling of dynamic cerebral autoregulation. Although the success of these techniques a number of weak points concerning dynamic regression tasks remain to be solved. The aim will then be to discuss how NARMAX type NN and SVM  models can be appropiately trained for this kind of dynamic processes.

Wed 16 - Thu 17 Sep-09 2-day Symposium Open and Interconnected Systems Modeling and Control
Site Oud Sint Jan, Brugge
9:00 am-5:00 pm
Organized by:
K.U.Leuven, ESAT-SCD (Signals, Identification, System Theory and Automation)

Venue: Site Oud Sint Jan, Brugge Belgium

Dates: Wednesday 16-Thursday 17 Sept 2009
Start at 9:00am until 5:00pm, Brugge Belgium
Conference website: http://www.esat.kuleuven.be/scd/oismc/

Mon 14 - Fri 18 Sep-09 14th Belgian-French-German CONFERENCE on Optimization (BFG 09)
Leuven
9:00 am-6:00 pm
Conference website: http://www.kuleuven.be/bfg09

Wed 2 - Wed 2 Sep-09 K.U. Leuven Seminars on Optimization in Engineering - Joachim Ferreau, Boris Houska
Esat Auditorium A
4:00 pm-5:30 pm
Boris Houska: "Solving Optimal Control Problems with ACADO Toolkit"

ACADO Toolkit is an open-source software environment and algorithm
collection for automatic control and dynamic optimization that is currently developed within OPTEC. It provides a general framework for using a great variety of algorithms for direct optimal control, including model predictive control and state/parameter estimation. ACADO Toolkit is implemented as self-contained C++ code providing a very user-friendly syntax for setting up optimal control problems. The object-oriented design allows for convenient coupling of existing optimization packages and for extending it with user-written optimization routines. In this talk we discuss several of ACADO Toolkit's key features and demonstrate how it can be used to solve optimal control problems. Also the underlying mathematical theory is briefly explained.

Hans Joachim Ferreau: "Using ACADO Toolkit for Parameter Estimation and Model Predictive Control"

Building on the introduction given by Boris Houska, this talk introduces
further functionality provided by ACADO Toolkit. We start with demonstrating how parameter estimation problems can be conveniently solved and briefly sketch the underlying Gauss-Newton SQP method that is employed by ACADO Toolkit. Besides its offline optimal control capabilities, ACADO Toolkit also provides a simulation environment to setup and test feedback controllers. We discuss how model predictive control problems can be formulated and show a couple of basic features to run closed-loop simulations.

Wed 19 - Wed 19 Aug-09 K.U. Leuven Seminars on Optimization in Engineering - Pieter Abbeel
ESAT Aud.A
4:00 pm-5:00 pm
"Apprenticeship Learning for Robotic Control with Application to Quadruped Locomotion and Autonomous Helicopter Flight"

Pieter Abbeel, UC Berkeley

ABSTRACT:  Slides

Many problems in robotics have unknown, stochastic, high-dimensional, and highly non-linear dynamics, and offer significant challenges to classical control methods.  Some of the key difficulties in these problems are that (i) It is often hard to write down, in closed form, a formal specification of the control task (for example, what is the objective function for "flying well"?), (ii) It is difficult to build a good dynamics model because of both data collection and data modeling challenges (similar to the "exploration problem" in reinforcement learning), and (iii) It is expensive to find closed-loop controllers for high dimensional, stochastic domains.  In this talk, I will present learning algorithms which show that these problems can be efficiently addressed in the apprenticeship learning setting---the setting when expert demonstrations of the task are available.  I will also present how our apprenticeship learning techniques have enabled us to solve real-world control problems that could not be solved before: They have enabled a quadruped robot to traverse challenging terrain, and a helicopter to perform by far the most challenging aerobatic maneuvers performed by any autonomous helicopter to date, including maneuvers such as chaos and tic-tocs, which only exceptional expert human pilots can fly.

BIO:

Pieter Abbeel received his Ph.D. degree in Computer Science from Stanford University in 2008. He is now an assistant professor at UC Berkeley's EECS department.  His research focuses on robotics, machine learning and control. For more information, see www.cs.berkeley.edu/~pabbeel

Wed 12 - Wed 12 Aug-09 ESAT Seminar - Leon Chua
Auditorium of the Arenberg Castle
2:00 pm-4:00 pm

Exciting lecture by Prof Chua, Unversity of California, Berkeley, the inventor of the Memristor, the missing element in circuit theory:

Lecture: Auditorium of the Arenbergkasteel, Kasteelpark Arenberg, 1,
Katholieke Universiteit Leuven, 3001 Heverlee

Title : "Memristors: From Proust to  HP"

Time: Wednesday August 12 2009 from 14u (till 16u)  (followed by a
coffee break in the salons)

Some background:
EVER had the feeling something is missing? If so, you're in good
company. Dmitri Mendeleev did in 1869 when he noticed four gaps in his
periodic table. They turned out to be the undiscovered elements
scandium, gallium, technetium and germanium. Paul Dirac did in 1929when he looked deep into the quantum-mechanical equation he had formulated to describe the electron. Besides the electron, he saw something else that looked rather like it, but different. It was only in 1932, when the electron's antimatter sibling, the positron, was sighted in cosmic rays that such a thing was found to exist.

In 1971, Leon Chua had that feeling. A young electronics engineer with a penchant for mathematics at the University of California, Berkeley, he was fascinated by the fact that electronics had no rigorous mathematical foundation. So like any diligent scientist, he set about trying to derive one.
And he found something missing: a fourth basic circuit element besides the standard trio of resistor, capacitor and inductor. Chua dubbed it
the "memristor". The only problem was that as far as Chua or anyone else
could see, memristors did not actually exist. Except that they do. Within the past couple of years, memristors have morphed from obscure jargon into one of the hottest properties in physics. They've not only been made, but their unique capabilities might revolutionise consumer electronics. More than that, though, along with completing the jigsaw of electronics, they might solve the puzzle of how nature makes that most delicate and powerful of computers - the brain. That would be a fitting pay-off for a story which, in its beginnings, is a triumph of pure logic.

Memristors:

(contact person for this event: Prof. Joos Vandewalle - K.U. Leuven ESAT-SCD)

Thu 23 - Thu 23 Jul-09 K.U. Leuven Seminars on Optimization in Engineering - Michael Saunders
Celestijnenlaan 200A - Auditorium 00.225
11:00 am-12:00 pm
"40 Years of Linear Algebra and Optimization at Stanford"
Michael Saunders, Stanford University

Abstract:

I came to Stanford in 1967 as a very green graduate student
(not in today's ecological sense).  Computer Science was a new
department, as was Operations Research.  The CS qualifying
exams allowed 3 out of 5 topics, including numerical analysis.
Alan George and I obtained permission to take one of the OR
exams.  Thus began a career of applying stable matrix methods
to numerical optimization (as pioneered by Gene Golub, Richard
Bartels, Philip Gill, and Walter Murray).

We trace the impact of Gene inviting numerous researchers to
Serra House (including Chris Paige and Bruce Murtagh), as well
as George Dantzig's creation of the Systems Optimization Lab
in the OR Department, and Gene's founding of the SCCM Program.

The talk includes some illustrations of the use of optimization
within the aerospace industry.

Bio:

Michael Saunders is a Research Professor in the Systems
Optimization Laboratory at Stanford University.  He obtained
his PhD in Computer Science at Stanford in 1972 (advisor Gene
Golub).  He is known for his contributions to software for
sparse linear equations (SYMMLQ, MINRES, LSQR, LUSOL) and
various optimization solvers (LSSOL, MINOS, NPSOL, PDCO, QPOPT,
SNOPT, SQOPT).  He teaches a class on Large-scale Numerical
Optimization.  He was elected Hon FRSNZ in 2007.

Mon 6 - Tue 7 Jul-09 Course on Algorithms for Convex Optimization - Lieven Vandenberghe
Computer Science 05.152
2:00 pm-5:00 pm
"Algorithms for convex optimization"
Lieven Vandenberghe (UCLA)

The lectures will give an introduction to some recent algorithms for
convex optimization.  In particular, we will cover two topics:

(1) Primal-dual interior-point methods for cone programming (linear,
second-order cone, and semidefinite programming), as they are implemented
in general-purpose software packages.

(2) Accelerated gradient algorithms and related first-order methods for
large-scale convex optimization.

Slides

Mon 29 - Mon 29 Jun-09 K.U. Leuven Seminars on Optimization in Engineering - Le Dung Muu
ESAT Aud. A
4:00 pm-5:00 pm
"On Mixed Variational Inequalities Involving D.C. Function: Application to a Nash-Cournot Market Equilibrium Model with Concave Cost Function"

prof. Le Dung Muu (VAST, Hanoi, Vietnam)
http://www.math.ac.vn/optimization/ldmuu/ldmuu.html

Abstract

Mon 22 - Mon 22 Jun-09 K.U. Leuven Seminars on Optimization in Engineering - Attila Kozma
ESAT 01.60
4:00 pm-5:00 pm
"Interval Analysis and Its Applications"
Attila Kozma, University of Szeged, Hungary

In my talk I intend to give an introduction to the world of intervals and its applications. Interval-based methods are widely used, among others for verifying the results of heuristic algorithms, theorem proof, solving differential equations and even for global optmization.
Firstly I would like to detail what I dealed with in my master thesis. Its aim was to solve a 2D circle packing problem with an approximation algorithm and to validate its results in a way. The different circle packing configurations can be represented by a system of nonlinear equations. This system ought to be solved by a method that gives a verified inclusion of the solution vector, if one exists. For this purpose I used the interval variant of Newton’s method for multivariate systems, which gives information about the existence and uniqueness of the solution.
Secondly a global optimization method will be presented that uses the inclusion functions. A Branch-and-Bound algorithm can be built to determine the global minima of an n-variable function on a certain box of its domain. Accelerating tests will be shown to make the exponential algorithm faster.

Slides

Wed 17 - Wed 17 Jun-09 SISTA Seminar - Kris De Brabanter
ESAT 00.57
4:00 pm
"Least Squares Support Vector Machines for large data sets: A Fixed Size Approach"
Kris De Brabanter (K.U. Leuven, ESAT-SCD)

The analysis of large scale data becomes more and more important in today's research. Hence, finding suitable methods capable of handling these large numbers of data is a challenge. In this presentation we will extend the Least Squares Support Vector Machines (LS-SVM) methodology so it can deal with these data sets. Our methodology is based on selecting a number of support vectors (a subset of the total number) using an entropy criterion. Based on these support vectors the mapping to the feature space can be approximated and the complete problem can be solved in the primal space. We provide some examples showing performance properties of the proposed method for the regression, classification and system identification case.

Tue 16 - Tue 16 Jun-09 SISTA Seminar - Harry Trentelman
ESAT 00.57
3:00 pm
"Optimal Robust Stabilization in Behavioral Framework"
Harry Trentelman (Univ. Groningen)

Given a nominal control system, together with a fixed neighborhood of this system, the problem of robust stabilization is to find a controller that stabilizes all systems in that neighborhood (in an appropriate sense). If a controller achieves this design objective, we say that it robustly stabilizes the nominal system. In this paper we formulate the robust stabilization problem in a behavioral framework, with control as interconnection. We use both rational as well as polynomial representations for the behaviors under consideration. We obtain necessary and sufficient conditions for the existence of robustly stabilizing controllers using the theory of dissipative systems. We will also find a smallest upper bound  on the radii of the neighborhoods for which there exists a robustly stabilizing controller. This smallest upper bound is expressed in terms of certain storage functions associated with the nominal control system.

Fri 12 - Fri 12 Jun-09 SISTA Seminar - Huseyin Akcay
ESAT 02.54
10:00 am
"Subspace-based rational interpolation of analytic functions from phase data"
Huseyin Akcay    (Center for Systems Engineering and Applied Mechanics (CESAME), Catholic University of Louvain)

(on leave from the Department of Electrical  and Electronics Engineering, Anadolu University,  Turkey)

In this talk, two simple subspace-based identification algorithms to identify
stable linear-time-invariant systems from corrupted phase samples of the
frequency response function are developed. The first algorithm uses data
sampled at non-uniformly spaced frequencies and is strongly consistent if
the corruptions are zero-mean and additive random variables with a
known covariance function. However, this algorithm is biased when the
corruptions are multiplicative, yet it exactly retrieves finite-dimensional
systems from noise-free phase data using a finite amount of data. The
second algorithm uses phase data sampled at equidistantly spaced
frequencies and also has the same interpolation property. A modified
implementation of this algorithm is strongly consistent if the corruptions
are zero-mean, additive random variables. This consistency property holds
for the multiplicative noise models as well provided that some noise
statistics are known a priori. Promising results are obtained when the
algorithms are applied to simulated data.

Slides

Wed 10 - Wed 10 Jun-09 SISTA Seminar - Tillmann Falck
ESAT 00.57
4:00 pm
"Using structure in LS-SVMs for the identification of Wiener-Hammerstein systems"
Tillmann Falck (K.U. Leuven, ESAT-SCD)

Least-Squares Support Vector Machines (LS-SVMs) are a powerful kernel based nonlinear regression technique. Wiener-Hammerstein systems are structured nonlinear systems consisting of a cascade of a linear dynamical system, a static nonlinearity and another dynamical system. It receives ongoing attention, for recent examples see the special benchmark session on Wiener-Hammerstein identification at SYSID09. The structure of the system can be seen as prior information and then be used to reduce the search space of the estimation problem. By including the structural information the demands on the training data are reduced and the prediction performance can be increased, both with respect to a full black box identification. To solve the problem in a convex fashion the model is overparametrized and then projected onto a related model class. The solution to the problem is given by the dual, but can also be obtained in the primal by approximating the feature map. This is advisable for large data sets as all data can be used to estimate a sparse model. Finally we will show results on some toy examples and the SYSID09 benchmark set.

Fri 5 - Fri 5 Jun-09 SISTA Seminar - Jorge Lopez
ESAT 00.57
2:00 pm
"Links between geometric, L1 and L2-SVMs"

In this seminar we will explore some results that we have derived in our department at Universidad Autónoma de Madrid, which stem from the well-known relationship between minimizing the distance between points in convex hulls (Convex Hull Nearest Point Problem - CH-NPP) and traditional SVMs with linear (L1-SVMs) and quadratic penalty terms (L2-SVMs). In particular, we will see the following: 1) how one of the best-known decomposition algorithms (SMO) and one of the most efficient geometric ones (MDM) are equivalent for L2 and convex hulls respectively, 2) a novel MDM algorithm for reduced convex hulls (RCH-MDM) deriving from L1-SMO, 3) an acceleration of CH-MDM based on the geometric observation that zigzags are present, which also works for L1-SMO, 4) an alternative proof of the asymptotic convergence of L2-SMO deriving from the CH-MDM convergence proof.

Wed 27 - Wed 27 May-09 K.U. Leuven Seminars on Optimization in Engineering - Brett Stewart
ESAT Aud. A
4:00 pm-5:00 pm
"Cooperative Model Predictive Control with Resource Management to Address Coupled Constraints"

Abstract:

In chemical plants, production often involves a series of unit operations interconnected through material and energy flows. Traditionally, in plants for which centralized control is judged impractical or unmanageable, the unit operations are controlled in a decentralized way so that closed-loop interactions between controllers are neglected.  It is well known, however, that if these interactions are strong plantwide performance is poor.

Cooperative model predictive control (MPC) has been proposed as a method for coordinating multiple optimization-based controllers.  In this control strategy, input trajectories are passed between controllers, each optimizing a plantwide objective.  It has been shown that cooperative MPC is provably stable and is plantwide optimal at convergence.  A necessary assumption for optimality is that each input is constrained independently, so that the plantwide feasible space is a Cartesian product of the feasible subspaces.  This assumption is not valid, however, for plants in which the constraints between controllers are coupled.  This situation arises if a common resource must be optimally shared between unit operations.

We propose an auxiliary optimization that enhances cooperative MPC to manage coupled constraints.  This resource manager decouples these constraints by computing the optimal inner hyperbox contained within the plantwide feasible space.  The optimization proceeds asynchronously and can be terminated at a suboptimal iterate.  We show the augmented optimization does not weaken the cooperative MPC stability properties and is plantwide optimal at convergence.  We provide some examples showing performance properties.

Slides

Tue 26 - Tue 26 May-09 K.U. Leuven Seminars on Optimization in Engineering - Mario Milanese
ESAT Aud. A
4:00 pm-5:00 pm
"Controlled airfoils for vessel on-board energy production"
Mario Milanese and Lorenzo Fagiano, Politecnico di Torino

Abstract:

This seminar illustrates the EU project KitVes - "Airfoil-based solution for Vessel on-board energy production" that aims to investigate the use of high-altitude wind energy to supply electricity for medium-to-large size boats. The basic idea is to exploit the aerodynamic forces of controlled tethered airfoils to generate electricity on-board, using suitable rotating mechanisms and electric generators placed on the ship. This way, it is possible to abate the consumption of fossil energy usually needed to supply electricity for the different on-board loads. Moreover, the generated energy can be used for traction purposes on electric-powered vessels, thus providing a completely "green" naval transportation system. After reviewing the innovative concept of high-altitude energy using tethered airfoils, the objectives and expected results of KitVes project will be described. The project participants are the companies Sequoia Automation (IT), Cesi (IT), Fatronik (ES), SVMtec (GER), Teks (FR), the University of Sheffield (UK), the Katholieke Universiteit Leuven (BE), the University of Wuppertal (GER), the Haute Ecole ARC (FR) and the company Modelway, spin-off of Politecnico di Torino (IT).

Slides

Wed 20 - Wed 20 May-09 Doctoral Presentation - Carlos Alzate
Wolfspoort Auditorium 00.08, Huis Bethlehem, Schapenstraat 34, 3000 Leuven
6:00 pm
"Support Vector Methods for Unsupervised Learning"
Carlos Alzate (K.U. Leuven, ESAT-SCD)

In the context of unsupervised learning, elements such as extensions to out-of-sample data, regularized formulations, model selection criteria and applicability to large-scale data become key aspects. Starting from least squares support vector machine (LS-SVM) core formulations, we provide extensions of unsupervised learning methods towards the incorporation of advanced elements. The primal-dual optimization framework typical of LS-SVM formulations allows extensions of the core models by adding additional constraints to the primal problem or by changing the loss function. This family of kernel methods is formulated in terms of high dimensional feature spaces at the primal level. The model at the dual level is represented in terms of Mercer kernels.
In this thesis, we formulate extensions for classical unsupervised learning applications such as principal component analysis (PCA), canonical correlation analysis (CCA), independent component analysis (ICA) and spectral clustering. In the case of kernel PCA, a formulation with general underlying loss function is proposed. This formulation allows the incorporation of robustness and sparseness into kernel PCA by using a robust epsilon-insensitive loss function. For kernel CCA, a multivariate formulation is described where regularization appears naturally in the primal problem. This method can be also used a contrast function for ICA when universal kernels are used. In the context of clustering, we propose a multiway spectral clustering methodology with out-of-sample extensions based on weighted kernel PCA together with a model selection criterion exploiting the structure of the solutions. This clustering method is further extended towards developing sparse models and incorporating prior knowledge. Real-life applications considered in this thesis include image segmentation, image denoising, time series clustering, image demixing and scientiﬁc journal clustering.

Promotor: Prof. J. Suykens

Wed 20 - Wed 20 May-09 K.U. Leuven Seminars on Optimization in Engineering - Herbert De Gersem
ESAT Aud. A
3:00 pm-4:00 pm
"Mixed-Integer Nonlinear Optimization of Accelerator Components"
Herbert De Gersem, K.U. Leuven
joint work with Thomas Hemker, Oskar von Stryk and Thomas Weiland (T.U. Darmstadt)

The numerical optimization of accelerator components with respect to continuous parameters and in combination with electromagnetic field simulation is standard. When integer-valued parameters are involved, the optimization procedure becomes prohibitively expensive. In the talk, we describe a sequential surrogate optimization scheme and apply the method for the optimization of the coil blocks of a superconductive magnet, recently designed for the GSI Helmholtz Center for Heavy Ion Research.

Slides

Mon 18 - Mon 18 May-09 K.U. Leuven Seminars on Optimization in Engineering - Yanna Stamati
ESAT Aud. A
4:00 pm-5:00 pm
“Exchange of tear film in the gap between a contact lens and the eye”
Ioanna Stamati, Philips Applied Technologies- National Technical University of Athens.

Abstract:

Contact lenses are worn predominantly in order to correct for aberrations of the eye. Movement of the lens over the eye, for example during a blink motion of the eyelids or a change in the gaze direction, is detrimental to the optical performance and in general should be minimized. Clinical experience indicates that no lens movement initially provides good comfort. However, such a non-moving lens later causes discomfort through accumulation of debris, acidity, poor hygiene etc. A "healthy fit" lens is considered by practitioners to be one with adequate movement during blinking, because this enhances the exchange of the tear film between the eye and the lens.

During blinking the eyelids exert a varying pressure on the lens, which causes the tear film to be sucked into or squeezed out of the gap between the contact lens and the eye. The lateral motion of the lens causes an exchange of the tear film through the shearing motion of the lens.

In this talk an approach is presented to study the exchange of tear film in the gap between a contact lens and the eye. In this approach the “spherical curved” contact lens-eye geometry is simplified to a flat disc placed on an infinite flat plane with a thin tear film in between them. The normal force exerted by the upper eyelid on the lens is modeled as a transient moving normal force acting at the lens. The Reynolds equation (combination of the Navier-Stokes equation and the continuity equation), which is well known in lubrication theory, is used to describe the tear film flow, as the tear film thickness is much smaller than the dimensions in the plane of the lens.

A finite element method is used to solve the Reynolds equation. Results will be shown for the squeeze motion of the lens, the pressure build up in the tear film,the tear film flow and the tear film exchange during a blink.

Thu 14 - Thu 14 May-09 K.U. Leuven Seminars on Optimization in Engineering - Greet Vanden Berghe
ESAT Aud. A
1:00 pm-2:00 pm
"Real world combinatorial optimisation problems. A personnel rostering example"
Greet Vanden Berghe, KaHo Sint-Lieven

Abstract:

Operational research provides a variety of powerful methods to solve combinatorial optimisation problems.  Initially, general purpose exact methods such as linear and integer programming algorithms were developed. Despite their general character, the applicability of such approaches is limited due to the unacceptably long computation time for real world problems.  Many special purpose algorithms and heuristics for particular problems became available but their real world applicability is poor due to restricted modelling power. Since the 1990's, local search and different metaheuristic approaches have become popular solution techniques to address diverse problems. They only require a limited implementation or configuration effort. Similarly, the quite recently introduced hyperheuristics are general purpose, rather easy to implement and their performance is suited for real world applicability.

The talk covers a brief overview of ongoing research at KaHo Sint-Lieven with respect to real world combinatorial optimisation problems.  Personnel rostering serves as an example to introduce a generic domain model and a hyperheuristic approach for tackling real world instances.

Wed 29 - Wed 29 Apr-09 K.U. Leuven Seminars on Optimization in Engineering - Johannes Paefgen
Celestijnenlaan 200A, room 5.152
5:00 pm-6:00 pm
'Robust Optimal Control of Vessel Processing in Container Terminals'
Johannes Paefgen (ETH, Zurich)

Abstract:

With growth of maritime container traffic being a long-term trend, container terminal operators are continuously facing challenges in automation and optimization of logistic processes. Efficiency as well as reliability are necessary to handle an increasing cargo throughput and to prevail in competition. Improvement of planning methods plays a crucial role in this context. Container vessels reach the port on a weekly basis within an arrival window of given length, according to a fixed periodic schedule. Their processing requires an assignment of quay cranes and berth sections to vessels, and should be completed within a contractually agreed service time limit. This can be formulated as a stochastic dynamic resource allocation problem. In this thesis, an adaptive planning solution for vessel processing in a container terminal under uncertainty is developed. A Markov decision process model is proposed, and algorithms for derivation of its optimal control policy are given for different cost criteria. Furthermore, the resulting closed-loop system is analyzed with respect to robustness for non-modeled disturbances. Theoretical findings are supplemented with an implementation for testing purposes, for which computational results are provided.
The approach presented allows to solve realistically sized instances and yields a state-dependent control law in the form of a look-up table with regard to scheduled vessels. For non-modeled exceptions, an extension to this operational plan is given in terms of a prediction of resource availability over a finite horizon, leading to a subordinate model predictive controller.

Mon 27 - Mon 27 Apr-09 K.U. Leuven Seminars on Optimization in Engineering - Saartje Arnout
ESAT Aud. B
11:00 am-12:00 pm
"Multi-scale design optimization of complex structures"
Saartje Arnout, K.U.Leuven.

In modern architecture, there is a tendency towards complex structures, such as (space) trusses and continuous shells. This stimulates an increasing attention for the optimal design of both the shape and the dimensions of such structures.
For truss structures, a more economic design is achieved by minimizing the volume while constraining the stress and the displacements. The truss size optimization is rather fast, as the derivatives can be computed by adjoint techniques.  For the combination of shape and size optimization, a double loop is adopted, as there are optimal bar sizes for every shape.  Using gradient-based methods for the outer shape optimization loop is therefore not trivial. The use of global optimization methods is considered. For shell structures, the strain energy is minimized with the stress and displacements constrained.  However, the real construction can deviate from the model used in optimization due to geometric imperfections or a change in the surroundings of the construction. To avoid unexpected failure, optimization must take these uncertainties into account. Stochastic constraints are therefore added to the optimization. This is especially important for shell structures, as the sensitivity of these structures to buckling is influenced unfavourably by small imperfections.

Thu 16 - Thu 16 Apr-09 K.U. Leuven Seminars on Optimization in Engineering - Pierre Martinon
ESAT Aud. A
4:00 pm-5:00 pm
"A less problem dependent indirect approach for optimal trajectories with singular arcs: application to space launchers"
Pierre Martinon, CMAP, Ecole Polytechnique, INRIA-Futurs.
http://www.cmap.polytechnique.fr/~martinon

The talk discusses trajectory optimization for space launcher problems, using Pontryagin's Minimum Principle and an indirect shooting method. Due to aerodynamic forces (mainly drag), the flight may involve singular arcs, i.e. phases with a non maximal thrust. As the analytical expression of several physical parameters is not known, realistic models typically use tabular data (atmospheric pressure for instance). For indirect approaches, this hinders the usual method for computing the singular control, which involves higher order derivatives. We introduce an alternate, less problem dependent method for computing an approximation of the singular control. Numerical experiments are carried out for a typical Ariane 5 and a new launcher prototype.

Slides , movie

Tue 3 - Tue 3 Mar-09 K.U. Leuven Seminars on Optimization in Engineering - Paul Goulart
ESAT 00.62
11:00 am-12:00 pm
"Model Predictive Control of Uncertain Systems with Bounded l2 Gain"
Paul Goulart, Imperial College London.

The talk will address the problem of designing a control law for a constrained linear system with bounded disturbances that ensures constraint satisfaction over an infinite horizon, while also guaranteeing that the closed-loop system has bounded l2 gain. The proposed strategy is based on repeated calculation of finite horizon control policies parameterized as affine functions of past disturbances. The controller can be implemented by solving a convex optimization problem at each time step, and performs identically to H-infinity control when constraints are removed.

Slides

Tue 3 - Tue 3 Mar-09 K.U. Leuven Seminars on Optimization in Engineering - Rodolphe Sepulchre
ESAT 00.62
3:00 pm-4:00 pm
"Computing with low-rank positive semidefinite matrices: a geometric approach."
Rodolphe Sepulchre, ULg.

Positive definite matrices have become a fundamental object of modern computational engineering. They appear in many computational problems in the form of variables (convex programming, LMIS problems, Lyapunov equations), covariance matrices (signal processing), diffusion tensors (biomedical imaging), kernels (machine learning), to cite a few. The numerical complexity of associated algorithms is typically O(n3) in the problem size, which prevents their use in a growing number of large-scale problems. Working with low-rank approximations of such matrices is a typical solution to reduce the numerical  complexity. In this talk, we study the underlying geometry of the set of fixed-rank positive semidefinite matrices. We argue that right geometries are essential to support the development of efficient low-rank versions of many existing algorithms and we illustrate its use in several examples.

Thu 12 - Thu 12 Feb-09 K.U. Leuven Seminars on Optimization in Engineering - Inna Sharf
Mechanical Engineering, Seminar Room C (03.42)
4:30 pm-5:30 pm
"Dynamic Locomotion with a Wheeled-Legged Quadruped Robot"
Prof. Inna Sharf, McGill University

Abstract:
In this talk, we present an overview of the work carried out in the Mechatronic Locomotion Laboratory at McGill University on a quadruped robotic platform, PAW. This robot features four springy legs with rotary actuation at the hips and wheels mounted at the  distal ends of the legs. The robot was designed to explore hybrid modes of locomotion, where it makes use of both wheels and legs to achieve novel behaviours. As well, the robot’s simple construction allows PAW to exploit the dynamics of a mass-spring system to achieve gaits such as bounding, galloping and jumping. In this presentation, we will first describe the basic design of the robot, its sensing capabilities, the control strategy implemented on the robot and the dynamics model of the robot, the latter developed using MSC Adams and Simulink software. We will then discuss in some detail several modes of locomotion that have been developed on the robot over the past three years.
Specifically, we present results for inclined turning and sprawled breaking achieved with the robot and a detailed investigation of the bounding gait with blocked and actuated wheels. The most recent addition to the robot’s repertoire is a dynamic jump. We will discuss the main stages of the jumping process and how the parameters governing the jump were optimized using the genetic algorithm coupled with the results of ADAMS/Simulink co-simulation. A simplified analytical model of the robot is also presented and dimensional analysis is employed to gain insight into the geometric and inertial parameters affecting the jump. The talk will conclude with some insights into further development of legged locomotion with robotic platforms.

Thu 12 - Thu 12 Feb-09 K.U. Leuven Seminars on Optimization in Engineering - Sasa Rakovic
ESAT-00.92
3:00 pm-4:00 pm
"Set Dynamics, Fixed Points & Control Synthesis"
Sasa V. Rakovic, Imperial College London

Abstract:

The talk discusses the utilization of the set theoretic analysis in control synthesis and analysis under constraints and uncertainty.
A methodology employing set dynamics and utilizing classical ideas from dynamical systems theory is invoked in order to discuss the minimality and the maximality of invariant sets as well as some computationally tractable control synthesis methods. Theoretical aspects concerned with the minimality and the maximality of invariant sets utilize fixed points of adequate image and preimage mappings induced from the underlying system dynamics, constraints on system variables and the uncertainty. It is revealed that the minimal and the maximal invariant sets are solutions, with special properties, under adequate but natural assumptions, of particular fixed point set equations.  Properties of proposed fixed point set equations are also utilized to indicate the fragility of receding horizon control synthesis process with respect to arbitrarily small and feasible perturbations of ingredients of the underlying finite horizon optimal control problem.
The power of the set dynamics approach is also invoked to re--examine briefly the minimality and the maximality of invariant sets for linear discrete time systems and to devise an adequate modification of standard reachability and viability algorithms. The proposed modification, in turn, leads to a computationally tractable method for the calculation of the outer invariant approximations of the minimal invariant set as well as the inner invariant approximations of the maximal invariant set. Developed procedures are dual to each other and produce sequences of improving invariant approximations of the minimal and the maximal invariant set, that are, under adequate conditions, monotonic set sequences and converge, respectively, to the minimal and the maximal invariant set. We also provide an explicit formula for estimates of the Hausdorff distance between the underlying set iterates and the minimal and the maximal invariant sets. The set dynamics approach is also utilized to discuss, at the conceptual level, a simple tube receding horizon control synthesis method for a class of non-linear discrete time systems guaranteeing a-priori, under mild assumptions, strong system theoretic properties of the controlled, uncertain, dynamics.

Biographical Information:

Sasa V. Rakovic obtained the PhD degree in Control Theory from Imperial College London. His PhD thesis, entitled Robust Control of Constrained Discrete Time Systems: Characterization and Implementation'', was awarded the Eryl Cadwaladr Davies Prize as the best PhD thesis in the Electrical and Electronic Engineering Department of Imperial College London in 2005.
He held posts of a Research Associate at Imperial College London (2004--11/2006) and a Postdoctoral Researcher at the ETH Zurich (11/2006--11/2008). He is currently an Honorary Research Associate at the CPSE of Imperial College London. His main research interests lie within the areas of control synthesis and analysis and decision making under constraints and uncertainty.

Wed 28 - Wed 28 Jan-09 K.U. Leuven Seminars on Optimization in Engineering - Carlos Dorea
ESAT 00.62
5:00 pm-6:00 pm
"Output Feedback Set-Invariant Controllers for Constrained Linear Systems"
Carlos E. T. Dorea
Presently with K.U. Leuven, ESAT/SCD

The concept of set-invariance provides useful tools for the synthesis of controllers for constrained systems, in particular, for constrained model predictive control with stability guarantee. If a linear system is stabilizable, it is possible to construct a controlled-invariant set such that a suitable state feedback control law can be computed to enforce the constraints along the state trajectory. We recently proposed a solution for the dual problem: to compute a polyhedral set which is conditioned-invariant with respect to a full-order state observer, in the sense that the estimation error trajectory can be kept in this set by means of a suitable (possibly nonlinear) output injection.

In this talk, these concepts are used to design output feedback controllers for linear systems subject to state and control constraints, additive disturbances and measurement noise. We first present necessary and sufficient conditions under which a polyhedral set is controlled-invariant under output feedback. Then, we show how a dynamic output feedback compensator structure can guarantee constraint satisfaction: from a pair composed by a conditioned-invariant and a controlled-invariant polyhedron, it is possible to construct an output feedback controlled-invariant polyhedron in the extended (system plus compensator) state space. Then, based on the available measurements and on the state of the compensator, a control sequence can be computed which enforces the state constraints, as long as the initial state belongs to an admissible set, which can  be characterized as well. Differently from other approaches, the compensator is not based on a precomputed linear observer and is, therefore, likely to result in a larger set of admissible initial states.

Slides

Tue 13 - Tue 13 Jan-09 Simon Stevin Lecture on Optimization in Engineering - Martin P. Bendsoe
Auditorium of the Arenberg Castle
5:00 pm-6:00 pm
9th Simon Stevin Lecture on Optimization in Engineering

"Topology Optimization of Multiphysics Systems - Successes and Challenges"

Martin P. Bendsøe, (Technical University of Denmark)
E-Mail: M.P.Bendsoe@mat.dtu.dk,
Web:  www.mat.dtu.dk/people/M.P.Bendsoe/

Flyer, Poster

Abstract:

The field of topology design involves the use of a multitude of computational tools. These encompass geometry modeling, analysis methods (primarily FEM) and optimization algorithms. That optimization has to be performed means that these three ingredients have to be linked together in a rational manner and that all tools are geared to the task. This talk provides an overview of these aspects and on some of the modeling and computational challenges that we face. Emphasis is on the work involved in extending the technology to new application areas involving new and multiple types of physics.

Biographical Information:

Martin Philip Bendsoe received his MS degree from the University of Copenhagen and his PhD from the Technical University of Denmark (DTU). Currently he is a full professor and head of the department of mathematics at DTU. He is a member of the executive committee of the International Society of Structural and Multidisciplinary Optimization since 1996, of the executive committee of the Congress Committee of the International Union of Theoretical and Applied Mechanics since 2004, and, of the editorial boards of Control and Cybernetics, Structural and Multidisciplinary Optimization (SMO), International Journal for Numerical Methods in Engineering (IJNME), Communications in Numerical Methods in Engineering, and Computers and Structures.
Dr. Bendsoe received several awards, such as the STATOIL Award and the Villum Kann Rasmussens Annual Award. He is the author of several well-known text books on topology optimization and related topics.

The "Simon Stevin Lecture Series on Optimization in Engineering" is set up in order to promote optimization in engineering. For this aim, every quarter of the year an outstanding international scholar is invited to report on latest progress in the development of optimization algorithms and their applications in engineering.
Simon Stevin (1548-1620) was a Flemish mathematician and engineer. Among other, he helped to advance the use of decimal fractions, was the first to explain the tides by the attraction of the moon, and discovered the hydrostatic paradox. He made numerous inventions, among them a wind propelled carriage with sails, the "land yacht", which once impressed Prince Maurice of Orange as it moved faster than horses, in around 1600 on the beach between Scheveningen and Petten. Simon Stevin was fond of promoting the use of science in daily life and in craftmanship, and translated various mathematical terms into dutch. Among other, he introduced the dutch word for mathematics, "wiskunde".

Directly after this Simon Stevin Lecture, a little reception will be given at 18:00 in the salons of Arenberg Castle, to which all attendants of the lecture are most warmly welcome!

***** REGISTRATION ENCOURAGED *****
Please send an e-mail with the subject "STEVIN" to optec.secretariaat@esat.kuleuven.be if you intend to participate in the event. No obligation, just to help us getting an idea how many people plan to come.

This Stevin lecture is co-sponsored by ICCoS (Identification and Control of Complex Systems), a Scientific Research Network of the Research Foundation - Flanders (FWO-Vlaanderen).

Mon 12 - Mon 12 Jan-09 K.U. Leuven Seminars on Optimization in Engineering - David Anisi
ESAT Aud.B
4:00 pm-5:00 pm
"Cooperative Surveillance with Multiple Unmanned Ground Vehicles"
David Anisi, Royal Institute of Technology Sweden

This talk considers the problem of cooperative task- and path planning for
a number of surveillance Unmanned Ground Vehicles (UGVs), such that a
given polyhedral area is covered by the UGV sensors. Both the minimum time
and the
connectivity constrained formulations will be discussed.

Wed 17 - Wed 17 Dec-08 K.U. Leuven Seminars on Optimization in Engineering - Erik Verriest
ESAT Aud. A
4:00 pm-5:00 pm
"Nonlinear Balancing and Model Reduction"
Erik Verriest,

Abstract:
A method for generalizing balancing to nonlinear systems is sketched. It differs from nonlinear balancing as introduced by Scherpen in 1994, and is based upon three principles:
1) Balancing should be defined with respect to a nominal flow;
2) Only Gramians defined over small time intervals should be used in order to preserve the accuracy of the linear perturbation model and; 3) Linearization should commute with balancing, in the sense that the linearization of a globally balanced model should correspond to the balanced linearizedmodel in the original coordinates.
The first two principles lead to local balancing, but it is shown that an integrability condition generically provides an obstruction towards a notion of a  globally balanced realization in the strict sense. The information obtained by local balancing of a nonlinear system already provides a lot of useful information about the dominant dynamics of the system and the topology of the state space. To accomplish local balancing, two Riemannian metrics are specified: One models the local reachability properties and one models the local observability properties. In general these are incompatible, inducing a different global topology, and thus explaining the aforementioned obstruction. Locally, it still may be possible to match these up, and local balancing at a point P corresponds to bending and reshaping the manifolds without tearing so that near P there is a snug fit (osculating contact) between the induced manifolds. Unlike the linear case, sensitivity and reduced modeling must be local concepts, and lead at best to a hybrid reduced model with modes of different dimension.
Finally, the use of stochastic reduced models will be mentioned, introducing a notion of uncertainty equivalence.

Biography:
Erik I. Verriest received the degree of ‘Burgerlijk Electrotechnisch Ingenieur’ from the State University of Ghent, Ghent, Belgium and the M.Sc. and Ph.D. degrees from Stanford University.  He was with  the Control Systems Laboratory and the Hybrid Computation Centre, Ghent, Belgium in 1973-74.  He joined the faculty of Electrical and Computer Engineering at Georgia Tech in 1980, and spent three years at Georgia Tech Lorraine in Metz, France.  His  interests are in mathematical system theory, with focus on periodic and hybrid systems, delay – differential systems, model reduction for nonlinear systems, and control with communication constraints. He served on several IPC’s and is a member of the IFAC Committee on Linear Systems.  He is presently with ESAT, KULeuven, Belgium.

Wed 10 - Wed 10 Dec-08 K.U. Leuven Seminars on Optimization in Engineering - Jan Gall
ESAT 00.62
4:00 pm-5:00 pm
"On Central Receiver Solar Power Plants"
Jan Gall, RWTH Aachen

The replacement of fossil fuels by renewable energy is one of the most challenging tasks in energy development. Concentrating solar power (CSP) systems represent one promising technology in this field. These systems use mirrors or lenses to focus sunlight into a small beam and utilize this bundled light as heat source for downstream processes.

This talk introduces Central Receiver Power Plants as one upcoming technology for solar power generation. An introduction to this technology will be given including a presentation of the main control tasks in this kind of power plants. Subsequently, the modelling aspects for numerical simulation will be presented. The talk concludes with a proof that one major task in Central Receiver Systems (CRSs), finding an optimal solar flux distribution on the receiver, is an NP-hard optimization problem.

Tue 9 - Tue 9 Dec-08 K.U. Leuven Seminars on Optimization in Engineering - Andrei Herdt
ESAT 00.92
2:00 pm-3:00 pm
"Stability optimization of a humanoid robot"
Andrei Herdt

Equilibrated walking in the presence of perturbations constitutes one of the most challenging problems in the development of humanoid robots. In order to improve the robustness of their equilibrium, numerous stability criteria have been developed so far, but most of them are heavily heuristic and overly conservative.
First, we review  recent results from my master thesis that address the problem of stability analysis by means of convex optimization methods. A problem for ellipsoidal approximation of stable state-spaces via optimization over linear matrix inequalities will be introduced.
Second, we give an outlook of the follow-up PhD project on "Nonlinear Model Predictive Control of Humanoid Robots" starting in January 2009 in Japan, and the inspirations that I got for this project in Leuven.

Tue 2 - Tue 2 Dec-08 K.U. Leuven Seminars on Optimization in Engineering - Luis N. Vicente
ESAT 00.62
4:00 pm-5:00 pm
"A globally convergent primal-dual interior-point filter method for  nonlinear programming: new filter optimality measures and computational results"
Luis N. Vicente, University of Coimbra
http://www.mat.uc.pt/~lnv/

Abstract:
In this talk we modify the original primal-dual interior-point filter method proposed by Ulbrich, Ulbrich, and Vicente (Math. Prog, 2004) for the solution of nonlinear programming problems.
We introduce two new optimality filter entries based on the objective function, and thus better suited for the purposes of minimization, and propose conditions for using inexact Hessians. We show that the global convergence properties of the method remain true under such modifications.

We also introduce a new optimization solver for the solution of nonlinear programming problems, called ipfilter, based on our primal-dual interior-point filter approach. The numerical results reported show that ipfilter is competitive both in efficiency and robustness and can handle large instances.

This is joint work with Renata Silva, Michael Ulbrich, and Stefan Ulbrich.

Fri 28 - Fri 28 Nov-08 SISTA Seminar - Andrés Véjar
ESAT 01.57
2:30 pm
"Network Modeling and Nonlinear Data Analysis"
Andrés Véjar

This work shows mainly two study cases and their relationships. The first case is a predator-prey system extension to which was added diferent species and sizes for the analysis. It was modified to be linked with real data of the Chilean marine ecosystem. The ob jective of this work is to
define a prototype model for marine ecosystem management. The second case is a study of a nonlinear time series to make a prediction model.
The data used is the copper price time series. For the analysis, the data is compared with the Chua's circuit time series to illustrate the non-stationarity with a simply visual method. A time dependant prediction model is developed using the recurrence plot information and also with
an empirical election of a good observer related to the time indexes. The model is mapped into a neural network for prediction. These two works are the start point for other related researches: the continous modeling of bio-inspired metaheuristics (Marriage in Honeys Bees, MBO), the validity
of the trophic relations in a diferent context (human organizations) and the closure in metabolic networks.

[PDF]

Wed 26 - Wed 26 Nov-08 K.U. Leuven Seminars on Optimization in Engineering - Andreas Potschka
Celestijnenlaan 200A, room 00.0144
4:30 pm-5:30 pm
"Optimizing Periodic Adsorption Processes: A Newton-Picard Inexact SQP Method"
Andreas Potschka, IWR Heidelberg

Periodic Adsorption Processes like the Simulated Moving Bed process for the separation of solute chemicals can be described by partial differential equations. For purification on a production scale, time periodic operation is crucial. It is known that these periodic solutions can be efficiently calculated by the Newton-Picard method. We describe an inexact SQP method for the optimization of periodic solutions which is based on the Newton-Picard method. It facilitates simultaneous Newton-Picard iterations within the optimization. Theoretical and numerical convergence results as well as the limits of the proposed method will be presented.

Tue 25 - Tue 25 Nov-08 K.U. Leuven emeritus
In het Auditorium van het Arenbergkasteel, Kasteelpark Arenberg 1, 3001 Leuven
3:30 pm
It is not every day, month nor year that a professor of SISTA is going to retire.
Actually, over the last 25 years nobody ever retired in our research group.
But this year, we will have an emeritus, namely Andre Barbe, and the celebration will be on
November 25:
http://www.esat.kuleuven.be/scd/abarbe/
You are all kindly invited to this unique event !

Mon 24 - Mon 24 Nov-08 Graduate Course "INTRODUCTION TO DERIVATIVE-FREE OPTIMIZATION" - Luis N. Vicente
Esat
9:15 am-12:15 pm
More details on the course website

Thu 20 - Fri 21 Nov-08 ESAT-dagen 2008
ESAT
Wij nodigen u graag uit op 20 en 21 november op de opendeurdagen van het departement Elektrotechniek-ESAT van de K.U.Leuven.

Via een rondleiding in de laboratoria ontdekt u de laatste trends op het gebied van telecommunicatie, energie, multimedia, data mining, beeld- en spraakverwerking, biomedische toepassingen, sensoren, bio-informatica, cryptografie, enz.

Op vrijdag 21/11 vinden ook de inaugurale lezingen plaats van de nieuwe docenten uit het departement.
http://www.esat.kuleuven.be/esatdagen2008/

Wed 19 - Wed 19 Nov-08 Graduate Course "OPTIMAL CONTROL AND OPTIMIZATION" - Erik Verriest
Esat
2:00 pm-5:00 pm
More details on the course website

Tue 18 - Tue 18 Nov-08 SISTA Seminar - Jan Willems
ESAT 00.62
3:30 pm-4:30 pm

INTERCONNECTED SYSTEMS

Jan C. Willems (K.U. Leuven, ESAT-SISTA)

Some of the characteristics of modern engineering systems are that they are open, interconnected, modular, and dynamic. Open' to allow interaction with their environment and interconnected' because they consist of an architecture of subsystems. The aim of this presentation is to describe a system theoretic language that deals with such systems. The central concept is a graph with leaves, with subsystems in the vertices, interconnections in the edges, and leaves for terminals that allow the interconnected system to  interact with its surroundings.  The behavioral framework used differs from the classical signal flow graph approach in that it does not formalize a system as a relation between inputs and outputs, and that it does not view system interconnection as output-to-input assignment. Input/output thinking does not capture the interaction of physical systems very well. In the behavioral approach, a dynamical system is simply a family of trajectories, interconnection means variable sharing, and control is interconnection. The modeling methodology presented will be further contrasted with bond graphs, circuit diagrams, and DAEs.

Hardly any mathematics is required to follow this back-to-basics lecture. It contains more pictures than formulas and uses only elementary mathematical concepts illustrated by very simple physical examples.

Thu 6 - Thu 6 Nov-08 K.U. Leuven Seminars on Optimization in Engineering - Michalis Frangos
Celestijnenlaan 200A, Room 00.144
4:30 pm-5:30 pm
"Model reduction by rational interpolation & low rank approximate solutions to large-scale Sylvester equations"
Michalis Frangos (Imperial College, London)

Abstract:
Krylov subspace projection methods are increasingly used for model reduction of large-scale linear systems. This presentation talks about the derivation of a set of simple equations for the rational Arnoldi algorithm, an algorithm which belongs to the class of Krylov subspace methods. This set of equations is proven to be useful for adaptive rational interpolation, error analysis and it allows the parameterization of the reduced order system in terms of a free parameter. In this presentation we suggest some choices of the free parameter to enhance the properties of the reduced order system and we suggest the use of the rational Krylov subspace techniques for the solution to large-scale Sylvester equations.

Fri 24 - Fri 24 Oct-08 K.U. Leuven Seminars on Optimization in Engineering - Georgios Chasparis
ESAT 02.58
10:00 am-11:00 am
"Distributed Dynamic Reinforcement of Efficient Outcomes in Multiagent Coordination"
Georgios Chasparis, Georgia Tech

We focus on the problem of distributed multiagent coordination. When agents have access to limited information about the environment (possibly other agents) and learn (locally) what to play through repeated experimentation, convergence to desirable (global) equilibria (equilibrium selection) might be challenging. To deal with this problem, we introduce a learning adaptation method, similar to reinforcement learning techniques, accompanied with decision rules that are based on feedback control (dynamic reinforcement). This learning framework exploits transient phenomena of the dynamics (off-equilibrium behavior) to reinforce convergence to efficient outcomes when the induced stochastic process has multiple resting points (equilibria). In particular, it is shown that non-efficient outcomes can be destabilized when dynamic reinforcement is applied by even a single agent. The utility of the proposed framework is illustrated in coordination games and distributed network formation, where non-efficient resting points of the stochastic process can be destabilized. In the case of distributed network formation, which is of independent interest, we also illustrate the utility of the proposed learning adaptation method to incorporate multiple design criteria, usually met in topology control for ad-hoc networks, which can reinforce convergence to desirable outcomes.

Fri 17 - Fri 17 Oct-08 One-day Workshop on Synchronization in Complex Networks
ESAT 00.62
Date: Fr. Oct 17, 2008
Place: K.U. Leuven, Department ESAT

Organizers:
J. Suykens (K.U. Leuven)
G. Osipov (Nizhny Novgorod)

Program

- Morning session ESAT 00.62 -

09:25-09:30 Welcome by J. Suykens (K.U. Leuven)

09:30-10:15 "Synchronization in a network of coupled discrete-time hybrid Lurie-type systems with strange hyperbolic nonstationary attractor",
V. Belykh (Nizhny Novgorod)

10:15-10:30 Coffee break

10:30-11:00 "Huijgens Synchronization",
H. Nijmeijer (T.U. Eindhoven)

11:00-11:30 "A partial synchronization theorem",
A.Y. Pogromsky (T.U. Eindhoven)

11:30-12:00 "Complete synchronization in ensembles of chaotic elements with a time-varying coupling",
E. Pankratova (Nizhny Novgorod)

12:00-13:30 Lunch in Alma restaurant

- Afternoon session ESAT 01.57 -

13:30-14:00 "Synchronization and clustering in networks of pulse-coupled integrate-and-fire oscillators",
A. Mauroy, R. Sepulchre (Universite de Liege)

14:00-14:30 "Synchronisation properties of Kuramoto oscillators: influence of coupling delay and symmetry",
O. D'Huys (VUB)

14:30-15:00 "Synchronization in a model based on global coupling of cells undergoing metabolic oscillations",
D. Gonze (ULB)

15:00-15:30 Coffee break

15:30-16:00 "Synchronization of nonlinear systems with time-varying delay coupling",
T. Oguchi (Tokyo Metropolitan University)

16:00-16:30 "Sequential synchronous activity in neural networks",
M. Komarov (Nizhny Novgorod)

16:30-17:00 "Synchronization effects in mixed media of coupled oscillary, excitable and passive systems",
V. Petrov (Nizhny Novgorod)

Information for interested participants: attendance to the workshop is free. If you want to participate in the lunch, please contact our secretary Ida Tassens (Ida.Tassens@esat.kuleuven.be).

Route description:
http://www.esat.kuleuven.ac.be/info/route.en.php

Fri 17 - Fri 17 Oct-08 K.U. Leuven Seminars on Optimization in Engineering - Ivan Lule, Jef Vanlaer, Jan Van Dierdonck
PW 01.06 at CIT
2:00 pm-4:00 pm
"On the influence of dynamic temperature input, initial cell population and oxidative stress on microbial dynamics at elevated temperatures"
Ivan Lule

"Optimal monitoring and control of complex (bio)chemical processes by means of intelligent data mining techniques "
Jef Vanlaer

"Unraveling the impact of sludge age and load on membrane fouling in membrane bioreactors based on a thorough characterization of the activated sludge properties"
Jan Van Dierdonck

This series of 3 seminars starts at 14u00 and takes place Friday October 17th, in room PW 01.06 at CIT (de Croylaan 46).

Wed 15 - Wed 15 Oct-08 Simon Stevin Lecture on Optimization in Engineering - Aharon Ben-Tal
Auditorium of the Arenberg Castle
5:00 pm-6:00 pm
8th Simon Stevin Lecture on Optimization in Engineering

"Robust Solutions of Uncertainty Affected Conic Optimization Problems"
Aharon Ben-Tal (Technion-Israel Institute of Technology)

Flyer, Poster, slides

Abstract We survey the main developments in Robust Optimization (RO), a methodology, which is aimed at solving optimization problems (static and dynamic) affected by uncertainty. We focus primarily on issues of computational tractability of the robust counterparts emerging from conic optimization problems (linear, conic quadratic and semidefinite programming). We then show how results pertaining to the latter issues can be used to solve difficult chance constrained programs under partial stochastic information. Finally, we discuss the synthesis of uncertainty affected discrete-time linear control systems by the RO methodology and illustrate the results by treating a supply chain problem.

Biographical Information:

Prof. Ben-Tal's research work is mainly in the area of nonlinear optimization. He published over 100 papers and 2 books (a third is forthcoming). His theoretical work is concerned with extremum principles for problems in a general setting, with regard to the underlying decision space, and the underlying smoothness of the functionals. He was among the first to develop a comprehensive theory of second-order optimality conditions for nondifferential problems.

Prof. Ben-Tal is also involved in research in stochastic mathematical programming. He introduced the concept of entropic-penalty for problems with randomness in the constraints, and developed a duality theory which established a link between stochastic programming and the Expected Utility principle in economics. Recently he developed, together with Prof. A. Nemirovski, the Robust Optimization methodology. The focus of Prof. Ben-Tal's work in recent years is in computational methods for solving large-scale continuous optimization problems. The algorithms he develops are used in designing optimally complex engineering structures, water distribution networks, and techniques for medical image reconstruction. The above projects are carried out in the MINERVA Optimization Center, a 1 million euro endowed research center headed by Prof. Ben-Tal.

Prof. Ben-Tal was a member of the International Council of the Mathematical Programming Society. He served as Area Editor of the journal Mathematics of Operations Research, and Associae Editor for Math. Programming and SIAM Optimization.He is currently a member of the Editorial Board of the journals Convex Analysis,J.Optimization and Engineering and European J. of Oprations Research.. He received Awards of Excellence from the Technion both for research and for teaching. In 2007 Prof. Ben-Tal was awarded the EURO Gold Medal-the highesr distinction for Operations Research within Europe.

The "Simon Stevin Lecture Series on Optimization in Engineering" is set up in order to promote optimization in engineering. For this aim, every quarter of the year an outstanding international scholar is invited to report on latest progress in the development of optimization algorithms and their applications in engineering.
Simon Stevin (1548-1620) was a Flemish mathematician and engineer. Among other, he helped to advance the use of decimal fractions, was the first to explain the tides by the attraction of the moon, and discovered the hydrostatic paradox. He made numerous inventions, among them a wind propelled carriage with sails, the "land yacht", which once impressed Prince Maurice of Orange as it moved faster than horses, in around 1600 on the beach between Scheveningen and Petten. Simon Stevin was fond of promoting the use of science in daily life and in craftmanship, and translated various mathematical terms into dutch. Among other, he introduced the dutch word for mathematics, "wiskunde".

Directly after this autumn's Simon Stevin Lecture, a little reception will be given at 18:00 in the salons of Arenberg Castle, to which all attendants of the lecture are most warmly welcome!

***** REGISTRATION ENCOURAGED *****
Please send an e-mail with the subject "STEVIN" to optec.secretariaat@esat.kuleuven.be if you intend to participate in the event. No obligation, just to help us getting an idea how many people plan to come.

This Stevin lecture is co-sponsored by ICCoS (Identification and Control of Complex Systems), a Scientific Research Network of the Research Foundation - Flanders (FWO-Vlaanderen).

Wed 15 - Wed 15 Oct-08 Financial guidelines, ICT and H.R. - SISTA
00.62
10:00 am-11:30 am
For all SISTA members:

H.R. in K.U.Leuven

How to spend money in a correct way?
Ordering things?
Getting reimbursed?
Using a SISTA credit card?

How to get a SISTA PC?
Where to get a Software License?
Where to announce a workshop?
How to get a website for your project?

Due to ever changing procedures we encourage all SISTIANS to attend this seminar
with coffee and cake!

to subscribe send a mail to Ida. ida.tassens@esat.kuleuven.be

Wed 15 - Wed 15 Oct-08 CIT Presentation to Ben-Tal and OPTEC members
CIT PW 00.38
2:00 pm-3:30 pm
"CIT Presentation to Ben-Tal and OPTEC members"

CIT, room: PW 00.38

Time schedule: 14h-15.30h .
- Welcome and introduction to BioTeC and CIT
- Presentation of biochemical engineering research and visit to bioreactors and predictive microbiology lab
- Presentation of waste water research and visit to waste water lab
- Presentation of chemical engineering research

Tue 14 - Tue 14 Oct-08 PMA Presentation to Ben-Tal and OPTEC Members
PMA 02.061
10:00 am-12:00 pm
PMA Presentation to Ben-Tal and OPTEC Members
"Rode zaal" (room 02.061), MECH-PMA
Tuesday October 14, 10:00-12:30

10u00: Jan Swevers: PMA Presentation
10u15: Myriam Verschuere: balancing
10u45: Lieboud Van den Broeck: Predictive Prefilter
11u15: Diederik Verscheure: Convex Time Optimal Robot Control
11u45: Demo of Time Optimal Robot Control

Thu 9 - Thu 9 Oct-08 K.U. Leuven Seminars on Optimization in Engineering - Erik Verriest
ESAT 00.62
4:00 pm-5:00 pm
"Maximally Smooth Transfers: Gluskabi Raccordations"
Erik I. Verriest

Abstract:
Smooth transfers for signals or trajectories of dynamical systems between elements of a particular subset of solutions are considered.  These can be stationary solutions, limit cycles, or more general classes.  The problem is first presented in a clear but abstract framework that sheds light on the constraints and the reason for imposing these constraints of the connecting solution (named Gluskabi raccordations after a mythical figure).  The solution is formulated as an optimal path in signal space. Then we look at specific interesting problems: the quasi-static transfers, known to thermodynamics, and their applications in static regime transfers of systems, orbit transfers in periodic regimes, e.g., of interest for smooth transitions between walking and running, and smooth transitions in harmonic signals (transients in the electric power industry). We also show a connection to the resolution of a paradox in optimal control.

Biography:
Erik I. Verriest received the degree of ‘Burgerlijk Electrotechnisch Ingenieur’ from the State University of Ghent, Ghent, Belgium and the M.Sc. and Ph.D. degrees from Stanford University.  He was with  the Control Systems Laboratory and the Hybrid Computation Centre, Ghent, Belgium in 1973-74.  He joined the faculty of Electrical and Computer Engineering at Georgia Tech in 1980, and spent three years at Georgia Tech Lorraine in Metz, France.  His  interests are in mathematical system theory, with focus on periodic and hybrid systems, delay – differential systems, model reduction for nonlinear systems, and control with communication constraints. He served on several IPC’s and is a member of the IFAC Committee on Linear Systems.  He is presently with ESAT, KULeuven, Belgium.

Thu 9 - Thu 9 Oct-08 K.U. Leuven Seminars on Optimization in Engineering - Thomas Wiese
ESAT Aud. A
3:00 pm-4:00 pm
"MPC Infeasibility Handling"
Thomas Wiese, TU Munich

Abstract:
After a short introduction to linear MPC we will discuss how one of the main advantages of MPC - consideration of hard constraints - can result in an infeasible problem. Different strategies for resolving the infeasibility will be presented and compared. Finally, an idea for an efficient implementation of a strategy into qpOASES will be outlined.

Slides

Fri 3 - Fri 3 Oct-08 Seminar - Bart Vandereycken
Celestijnenlaan 200A, Auditorium 00.225
11:00 am-12:00 pm
"Solving Lyapunov equations by geometric optimization on the  manifold of low-rank matrices"
Bart Vandereycken, Department of Computer Science, TWR

Place: Celestijnenlaan 200A, Auditorium 00.225, 3001 Leuven
Date and Time: Friday, October 3, 2008,  11h-12h

Abstract:

We present a geometric  optimization approach to approximate solutions of certain  matrix equations, e.g. the Lyapunov  equation, by low-rank positive semidefinite (PSD) matrices.

The solution of  a matrix equation is in general  a dense matrix which poses significant problems in  large-scale applications. The mere fact that the number of unknowns scales quadratically with the problem size necessitates  some sort  of approximation  of the  solution  itself. A popular  technique  to overcome  this  curse  of  dimensionality is  a low-rank approximation. We  will  obtain  these  approximations  by  minimizing  an  objective function  representing  the  error  defined  on the  set  of  low-rank matrices.  Because  this set is a  smooth manifold, we end  up with an unconstrained minimization  problem  on  a  Riemannian  manifold.  An advantage of  this approach is that  we have reduced  the dimension of the original problem to the dimension of the manifold, which is linear in the number of discretization variables.

However, standard optimization  techniques are not directly applicable since the problem  is no longer defined on  a Euclidean space. Luckily we  can exploit the  Riemannian geometry  of the  set of  low-rank PSD matrices by  using geometric optimization.  This leads to a  series of consecutive minimizations,  each defined on  the tangent space  of the current  approximation.  This way  we  retrieve a  predictor-corrector algorithm:
1) Make a step in the  tangent space by minimizing a suitable  model
of the  cost function.
2) Retract this step to the manifold, e.g. by projection.

This framework  includes well-known geometric  optimization techniques on manifolds,   like    non-linear   CG,   Riemannian   Newton   and Trust-Region. In our case this can also be related to the steady stateof  a gradient  flow  integrated  on the  manifold.  The geometry  and efficient implementation of the manifold as well as locally optimal CG and Trust-Region methods are discussed and illustrated numerically.

Tue 30 - Tue 30 Sep-08 SISTA Seminar - Leonid Mirkin
ESAT Aud. B
5:00 pm-6:00 pm
"Sampling Theorem from a System-Theoretic Viewpoint"

Leonid Mirkin
Faculty of Mechanical Engineering
Technion - Israel Institute of Technology
Haifa

Abstract: This talk will present a system-theoretic viewpoint on the
(classical) problem of reconstructing a continuous-time signal from
its samples. Namely, it is assumed that the continuous-time signal is
modeled as the output of a linear system and the reconstruction
problem is formulated as an optimal model-matching problem, in which
the performance is measured by system norms. This leads to an
alternative proof of the celebrated Sampling Theorem and some less
studied extensions. I'll try to keep the talk self-contained, so that
no prior knowledge of sampling theory is required.

Thu 25 - Thu 25 Sep-08 K.U. Leuven Seminars on Optimization in Engineering - Jakob Bjoernberg
ESAT 00.57
3:00 pm-4:00 pm
"Robust model predictive control for constrained, linear systems through approximate dynamic programming"
Jakob Björnberg,
University of Cambridge, England, and
Royal Institute of Technology, Stockholm, Sweden

Abstract:
The talk begins by reviewing how one can approach robust model predictive control for discrete-time, uncertain, constrained systems by dynamic programming. We then specialize to a certain class of linear systems with parametric uncertainties, so-called polyhedral dynamic programming, and demonstrate how to represent the cost-to-go functions and feasible sets exactly and compactly in terms of polyhedra in this case. As a method to lower the computational complexity we then present an approximation technique for dynamic programming that is suitable for this problem class. This is at the expense of optimality, but nevertheless allows to generate robustly stable feedback laws that are guaranteed to respect all constraints.

Tue 23 - Tue 23 Sep-08 Course on Numerical Optimization - Moritz Diehl
200 S.00.05
10:35 am-12:35 pm
Course on Numerical Optimization
Tuesdays and Wednesdays, 10:35 - 12:35,
Celestijnenlaan 200 S.00.05
Starting on September 23, 2008

Course announcement

Aim of the course (H03E3a ,6 credit points, 17 lectures, 7 exercise sessions) is to provide the
basics of numerical optimization methods and to enable the participants to apply and adapt
these methods to practical engineering problems. From October 24 on, the course will be
accompanied by weekly computer exercise sessions. A final written exam in January 2009 will
conclude the course.

Contents of the Course

Part I: Introduction
1. Fundamental Concepts of Optimization
2. Types of Optimization Problems
3. Convex Optimization, Linear Programming (LP)

Part II: Unconstrained Optimization
4. Estimation and Fitting Problems
5. Newton Type Optimization Methods
6. Gauss-Newton and Levenberg-Marquard
7. Calculating Derivatives
8. Trust Region vs. Line Search Methods

Part III: Constrained Optimization
9. Karush-Kuhn-Tucker Optimality Conditions
10. Parametric Sensitivity of Solutions
11. Quadratic Programming (QP): Active Set
12. Interior Point Methods
14. Globalisation Strategies
15. Optimization Modelling, Slack Variables
16. Optimal Control Problems
17. Summary of the Lecture

Lecture: Moritz Diehl, Professor for Optimization in Engineering,
ESAT 02.82, moritz.diehl@esat.kuleuven.be
Exercises: Boris Houska, Hans Joachim Ferreau
The course is partly based on the book Numerical Optimization by J.
Nocedal and S. Wright, Springer Verlag, and on the free book:
Convex Optimization by S. Boyd and L. Vandenberghe, Cambridge

Tue 23 - Tue 23 Sep-08 K.U. Leuven Seminars on Optimization in Engineering - Hoang Tran
Esat, 00.57
4:00 pm-4:30 pm
"An implementation of primal-dual interior point methods for solving Quadratic Programs"
Hoang Tran

Abstract:
This presentation describes the work which I have done during my time at K.U Leuven from July 1st to September 25th under the supervision of Professor Moritz Diehl , Mr Boris Houska and Mr Quoc Dinh Tran. Under their supervision, I have implemented a QP Solver using the Primal-Dual Interior Point Methods. In this presentation, we would like to begin by outlining the algorithm then proceed to some aspects of our implementation. Some examples and a brief description of the performance of the program for problems with large dimensions may also be included if time permits.

Slides

Thu 4 - Thu 4 Sep-08 K.U. Leuven Seminars on Optimization in Engineering - Toshiyuki Ohtsuka
ESAT Aud. A
11:00 am-12:00 pm
"Solutions to the Hamilton-Jacobi Equation with Algebraic Gradients"
Prof. Toshiyuki Ohtsuka (Osaka University)
http://www-sc.sys.es.osaka-u.ac.jp/~ohtsuka/

Abstract:
The Hamilton-Jacobi equation (HJE) is a fundamental equation in the analysis and control of nonlinear systems. However, the HJE cannot be solved explicitly in general. In this talk, the HJE with coefficients consisting of rational functions is considered, and its solutions with algebraic gradients are characterized in terms of commutative algebra. It is shown that there exists a solution with an algebraic gradient if and only if an ideal satisfying some conditions exists in a polynomial ring over the rational function field. If an appropriate ideal is found, the gradient of the solution is defined implicitly by a set of algebraic equations. Then, the gradient is determined by solving the set of algebraic equations pointwise without storing the solution over a domain in the state space. Thus, the so-called curse of dimensionality can be removed when a solution to the HJE with an algebraic gradient exists. A new class of explicit solutions for a nonlinear optimal regulator problem is given as an application of the present approach.

Wed 3 - Wed 3 Sep-08 K.U. Leuven Seminars on Optimization in Engineering - Fjo De Ridder
ESAT, Room 00.62
4:00 pm-5:00 pm
A distributed control algorithm for the electricity infrastructure

Fjo De Ridder, Flemish Institute for Technological Research
VITO - Energy Technology, http://www.vito.be
(Joint work with Maarten Hommelberg and Guy Vekemans)

Today we see the penetration of PV panels on our roofs and wind power in rural areas. In the near future combined heat and power systems will turn our homes into small electricity plants. This ongoing shift from central generation to decentral generation causes the need for radical changes in the electricity infrastructure.

The electricity infrastructure has up to now been controlled in a centralized, top-down structure. Decentralisation of power generation creates the need for decentralised - or distributed - control. New concepts will have to ensure that the power system remains secure.

The control algorithm, presented, is a software concept for distributed control of power producers, power consumers and storage systems. It enables control of a cluster of devices, such that the cluster behaves as one single system. This is achieved by tuning demand and supply within the cluster.

In this presentation, we will briefly discuss this distributed control algorithm, one or two applications, some of our projects and will highlight the most important shortcomings of the algorithm.

Slides

Thu 3 - Thu 3 Jul-08 K.U. Leuven Seminars on Optimization in Engineering - Nguyen Trung Thanh
Esat 00.62
4:00 pm-5:00 pm
"An inverse problem for partial differential equations with application in burried object detection"
Nguyen Trung Thanh, VUB.

This talk introduces an inverse problem for partial differential equations which arises in the detection and classification of buried objects using thermal infrared images of the ground. It consists of two main parts: thermal modeling and inverse problem. The aim of the thermal modeling is to simulate diurnal distribution of the soil temperature using a partial differential equation-based thermal model. Topics considered in this part are: (i) formulation of the thermal model; (ii) numerical methods for solving it; (iii) methods for estimating necessary input parameters of the model in practical situations using in situ measurements; and (iv) validation of the model for landmine detection by comparing simulations with outdoor experimental data. The inverse problem is aimed at detecting buried objects (anomalies) and classifying them by estimating their thermal and geometrical properties (shape, size). This part includes: (i) mathematical setup of the inverse problem using a least squares approach; (ii) its simplification for the case of cylindrical objects and a two-step method for solving it; (iii) gradient-based optimization algorithms with the formulation of the objective function’s gradient using a discrete adjoint method; and (iv) application in the detection of buried landmines.

Short CV:
Nguyen Trung Thành was born in Hatinh, Vietnam in 1980. He received the Bachelor and Master in Mathematics at Hanoi University of Science, Vietnam in 2002 and 2003, respectively, and PhD in Engineering at Vrije Universiteit Brussel, Belgium in 2007.
He is currently working as a postdoc researcher at department of Electronics and Informatics, Vrije Universiteit Brussel.
His main research interests involve: (i) inverse and optimal control problems of systems governed by partial differential equations (PDEs); (ii) optimization algorithms; (iii) numerical methods for partial differential equations; (iv) modeling and simulation; and (v) inverse scattering theory.

Mon 30 - Mon 30 Jun-08 OPTEC Seminar - Elias Jarlebring
Department of Computer Science, Room 00.147, Celestijnenlaan 200A
1:00 pm-2:00 pm
On numerical methods for the spectrum of delay-differential equations
Elias Jarlebring, T.U. Braunschweig

We consider delay-differential equations (DDEs) with constant coefficients and show some ways to analyze numerically properties of such equations. Many properties of such DDEs, e.g., stability,
oscillation and asymptotic behavior in general, can be determined from the spectrum, i.e., the solutions of the characteristic equation. The characteristic equation of a DDE is, unlike the characteristic equation for ordinary differential equations, a transcendental equation with an infinite number of solutions. We give an overview of the current numerical methods aiming to find roots of the characteristic equation.

Numerical methods to compute the delay margin, i.e., minimum delay which causes the DDE to turn unstable, are also discussed. In particular, we show that the critical delays, i.e., those valued of the  delays for which the DDE to have a purely imaginary eigenvalue, can be computed from the solutions of a polynomial eigenvalue problem.

Finally, we mention some examples where the current numerical methods do not work efficiently, such as problems of large dimension, DDEs with a large number delays or distributed delays and badly conditioned neutral and differential-algebraic problems.

Thu 19 - Fri 20 Jun-08 OPTEC Workshop on Distance Measures and Eigenvalue Optimization
Room 05.152, Department of Computer Science, Celestijnenlaan 200 A
OPTEC Workshop on Distance Measures and Eigenvalue Optimization

Aim:
The aim of the workshop is to bring together researchers interested in optimization problems involving eigenvalues, stability and robustness measures. It is at the intersection of numerical optimization (nonsmooth optimization, semi-definite programming), linear algebra (matrix distance problems, pseudospectra and stability radii) and application areas, with the emphasis on problems from systems and control (fixed-structure control design problems, design of open loop stable systems).

Invited speakers:
• Didier Henrion (LAAS-CNRS, France and Czech Technical University in Prague, Czech Republic)
• Katja Mombaur (University of Heidelberg and LAAS-CNRS, Toulouse, France)
• Dominikus Noll (Universite Paul Sabatier, Toulouse, France)
• Michael Overton (New York University, USA)
Program, reistration and practical details:
http://homes.esat.kuleuven.be/~optec/events/workshops/EIGOPT/index.php

This workshop is co-sponsored by ICCoS (Identification and Control of Complex Systems), a Scientific Research Network of the Research Foundation - Flanders (FWO-Vlaanderen).

Wed 18 - Wed 18 Jun-08 Simon Stevin Lecture on Optimization in Engineering - Michael Overton
Auditorium of the Arenberg Castle
5:00 pm-6:00 pm
7th Simon Stevin Lecture on Optimization in Engineering

"Nonsmooth, Nonconvex Optimization, with Applications in Control" Slides
Michael Overton
Professor of Computer Science and Mathematics (New York University)
http://www.cs.nyu.edu/overton/

Abstract:

There are many algorithms for minimization when the objective function is differentiable, convex, or has some other known structure, but few options when none of the above hold. We describe two simple algorithmic approaches for minimization of nonsmooth, nonconvex objectives: Gradient Sampling (a new method that, although computationally intensive, has a nice convergence theory, which we will outline), and BFGS (a new look at an old method, for which we have no theoretical results, but some interesting empirical observations).
Both methods require the user to provide a routine that computes the function value and, if it exists, the gradient at a given point, the idea being that the function is virtually never evaluated at a point where it is not differentiable, even though it is typically not differentiable at local optimizers.
Applications abound in control. Of particular interest to us is H-infinity fixed-order controller design. We briefly describe our code HIFOO: H-Infinity Fixed-Order Optimization, which is built on HANSO: Hybrid Algorithm for Non-Smooth Optimization.

Bibliographical Information:

Michael L. Overton received his BSc from UBC in 1974, along with the Governor General’s Gold Medal for Arts and Sciences. He received the MS and PhD degrees in Computer Science from Stanford University. He is currently Professor of Computer Science and Mathematics at the Courant Institute of Mathematical Sciences, New York University. Michael Overton is an elected member of the Board of Trustees of SIAM (Society for Industrial and Applied Mathematics) and has also served on the SIAM Council. He is a member of the Council of FoCM (Foundations of Computational Mathematics) and of the Board of Directors of the Canadian Mathematical Society. He serves on the editorial boards of SIAM Journal on Optimization (for which he was Editor-in-Chief from 1995-1999), SIAM Journal on Matrix Analysis and Applications, the IMA Journal on Numerical Analysis, and SIAM Review. His research interests are at the interface of optimization and linear algebra, especially nonsmooth optimization problems involving eigenvalues, with applications to many dierent subjects including robust control, structural analysis, combinatorial optimization and convex analysis. He is author of "Numerical Computing with IEEE Floating Point Arithmetic" (SIAM, 2001).

The "Simon Stevin Lecture Series on Optimization in Engineering" is set up in order to promote optimization in engineering. For this aim, every quarter of the year an outstanding international scholar is invited to report on latest progress in the development of optimization algorithms and their applications in engineering.
Simon Stevin (1548-1620) was a Flemish mathematician and engineer. Among other, he helped to advance the use of decimal fractions, was the first to explain the tides by the attraction of the moon, and discovered the hydrostatic paradox. He made numerous inventions, among them a wind propelled carriage with sails, the "land yacht", which once impressed Prince Maurice of Orange as it moved faster than horses, in around 1600 on the beach between Scheveningen and Petten. Simon Stevin was fond of promoting the use of science in daily life and in craftmanship, and translated various mathematical terms into dutch. Among other, he introduced the dutch word for mathematics, "wiskunde".

Directly after this Simon Stevin Lecture, a little reception will be given at 18:00 in the salons of Arenberg Castle, to which all attendants of the lecture are most warmly welcome!

***** REGISTRATION ENCOURAGED *****
Please send an e-mail with the subject "STEVIN" to Ida.Tassens@esat.kuleuven.be if you intend to participate in the event. No obligation, just to help us getting an idea how many people plan to come.

This Stevin lecture is co-sponsored by ICCoS (Identification and Control of Complex Systems), a Scientific Research Network of the Research Foundation - Flanders (FWO-Vlaanderen).

Wed 18 - Wed 18 Jun-08 K.U. Leuven Seminars on Optimization in Engineering - Romain Dujol
Esat Auditorium A
11:00 am-12:00 pm
Augmented Lagrangian algorithm for infinite-dimensional constrained optimization

In this talk, we will present an extension of the classical augmented Lagrangian algorithm to the infinite-dimensional framework designed by Ekkehard SACHS and Annick SARTENAER. We will show that the changes required for the extension are minimal. A discretized version of the algorithm is derived from the abstract extension.
The so-called discretized algorithm performs a sequence of finite-dimensional augmented Lagrangian subproblems with eventually dynamically determined discretization levels to ensure global convergence.
Some conceptual concerns that came up during the implementation of the algorithm are presented along with some preliminary results.

This talk will be precedeed by a brief review of the speaker's Ph.D research on man-made satellite optimal steering.

Tue 17 - Tue 17 Jun-08 K.U. Leuven Seminars on Optimization in Engineering - Carlo Savorgnan
Esat 00.62
5:00 pm-6:00 pm
"Nonlinear optimal control via occupation measures"
Carlo Savorgnan, LAAS, Toulouse

Abstract:
In this talk, we will consider nonlinear optimal control problems with state and input constraints. First, we will show how these problems can be reformulated in terms of occupation measures, in order to obtain an infinite dimensional linear programming (LP) problem. Second, in the special case when all problem data are polynomial, we will provide a hierarchy of linear matrix inequality relaxations of the LP problem. Finally, the solution to these relaxations will be used in order to compute an almost optimal control law.

Thu 12 - Thu 12 Jun-08 OPTEC Topical Workshop on Parameter Estimation in Differential Equations
ESAT AUD A
10:00 am-4:45 pm
Aim of the workshop is bring together all those people within OPTEC and its partners, that work on estimation of parameters with underlying differential equation models. Both, numerical methods and applications shall be covered. Special emphasis is put on applications from civil-, mechanical-, chemical-, and bio-engineering. The workshop will include a free sandwich lunch at 12:30.

Please register indicating if you participate in the lunch  with Julian.Bonilla@cit.kuleuven.be, latest before May 30 (the number of places is limited).

The  workshop is organized by Julian Bonilla and Moritz Diehl

Tue 27 - Tue 27 May-08 OPTEC Meeting on Dynamic Optimization Software
ESAT, Room 00.62
10:00 am-1:30 pm
OPTEC Meeting on Dynamic Optimization Software

Goals
1. Exchange knowledge regarding the different types of problems OPTEC members are ad-
dressing.
2. Create groups that address similar problems for instance nonlinear dynamic optimization
problems, LPs or MILP perhaps in different engineering fields.
3. Combine programming efforts for the ones who intend to program their own algorithms.
4. Discuss technical details regarding standards to set a common language
5. Get advices from the more trained OPTEC programmers among us.
6. Strengthen cooperation among OPTEC members.

The meeting will consist in 20 minutes presentations by PhD students about software development challenges.

A sandwich lunch is gratefully offered to participants

Wed 21 - Wed 21 May-08 OPTEC Seminar - Matthew Peet
Room A05.001, Celestijnenlaan 200A, Department of Computer Science
11:00 am-12:00 pm
Computation and Control: New Methods and New Possibilities
Matthew Peet, INRIA Rocquencourt

Abstract:

The study of the control systems is one of the oldest disciplines in engineering, yet recently there has been a substantial change in the way it is practiced. Powerful computers and efficient algorithms for optimization have replaced graphs and tables as the tools of the practicing control engineer. In fact, most problems that control engineers are currently working on are complex enough that devising a reliable controller would be a practical impossibility without computational aids.

In this talk, we give examples of the new types of complex systems which control engineers have begun to investigate in recent years. These examples come from Internet congestion control and cancer therapy and include features such as nonlinearity, delay, and decentralized structure. To address these systems, we adopt ideas from computer science and optimization and combine them with control theory using results from mathematical analysis. The results are used to design numerical algorithms which are then applied to real problems in Internet congestion control and cancer therapy.

Biography of the discussant

Matthew Peet received the B.S. in Physics and in Aerospace Engineering from the University of Texas at Austin in 1999 and the M.S. and Ph.D. in Aeronautics and Astronautics from Stanford University in 2001 and 2006. He is currently a Postdoctoral Fellow at the National Institute for Research in Computer Science and Control (INRIA) near Paris, France. His research interests include computational methods of analysis and control of nonlinear and distributed systems. He works extensively on polynomial optimization algorithms including the sum-of-squares methodology.

Fri 9 - Fri 9 May-08 OPTEC Seminar - Christopher Byrnes
ESAT 00.62
2:00 pm-3:00 pm
"IMPORTANT MOMENTS IN SYSTEMS AND CONTROL"
Christopher I. Byrnes
Department of Electrical and Computer Engineering
Washington University
St Louis, Missouri, USA

The moment problem as formulated by Krein and Nudel’man is a beautiful generalization
of several important classical moment problems, including the power moment problem,
the trigonometric moment problem and the moment problem arising in Nevanlinna-Pick
interpolation. Motivated by classical applications and examples, in both finite and infinite
dimensions, we recently formulated a new version of this problem that we call the moment
problem for positive rational measures. The formulation reflects the importance of rational
functions in signals, systems and control, where there have been applications to circuit
theory, model reduction, time-optimal control, robust control, signal processing, spectral
estimation, stochastic realization theory and the use of the moments of a probability density.
While the rational version of the problem is decidedly nonlinear, the basic tools still
rely on convexity. In this talk, we present a solution to this problem that combines basic
analysis and topology in Euclidean space with a nonlinear convex optimization problem
that generalizes the maximum entropy approach used in several classical special cases.

Short biography:
Chris Byrnes has made fundamental contributions to the application of mathematics to
engineering, particularly to the areas of systems, signals and control. The author of about
250 technical papers and books, he and A. Isidori received the IEEE George Axelby Best
Paper Award for their work on output regulation for nonlinear control systems in 1991. In
1993, they shared the IFAC Automatica Best Paper Prize for a paper in which they solved
a longstanding problem on the control of a rigid body model of a satellite in an actuator
failure mode. In 2003, T. T. Georgiou, A. Lindquist and Chris shared the IEEE George
Axelby Best Paper Award for work applying these methods to Nevanlinna-Pick
interpolation. In 2005, he was awarded the Reid prize from SIAM for his contributions to
Control Theory and Differential Equations. He was named the 2008 Hendrik Bode Prize
Lecturer by the IEEE and will hold the Giovanni Prodi Chair in Nonlinear Analysis at the
University of Wuerzburg in 2009. He is a Fellow of the IEEE and a Foreign Member of
the Royal Swedish Academy of Engineering Sciences. His current interests include the
application of dynamics and topology to nonequilibrium problems in control, particularly
the behavior of periodic steady state response for nonlinear control systems.

(slides  http://homes.esat.kuleuven.be/%7Eoptec/events/20080509_byrnes.pdf )

Mon 5 - Mon 5 May-08 SISTA Seminar - Bart Vanluyten
Auditorium of the Arenberg Castle
5:00 pm
"Realization, Identification and Filtering for Hidden Markov Models using Matrix Factorization Techniques"
Bart Vanluyten (K.U. Leuven, ESAT-SCD)

Abstract:      Since their introduction in 1957, hidden Markov models have been used in several engineering applications (speech processing, computational biology). However, many theoretical questions concerning hidden Markov models remain open until this moment. Contributing to these theoretical questions forms the first main objective of this thesis. When considering the theoretical problems, we find inspiration in the analogy with the corresponding problems for linear stochastic models. The solution to most of the problems concerning linear stochastic models makes use of the singular value decomposition. For the solution of the corresponding problems for hidden Markov models, it turns out that modifications to the nonnegative matrix factorization are needed. Investigating new nonnegative matrix factorization techniques forms the second main objective of this thesis.      A first theoretical problem concerning hidden Markov models is the exact positive realization problem. No procedure is known to solve this problem. In this thesis, two relaxed versions of the problem are solved: the exact quasi realization problem and the approximate positive realization problem. A second problem is the identification problem for hidden Markov models. In this thesis we propose an identification approach that estimates the state sequence directly from the output data and subsequently computes the system matrices from the obtained state sequence and the given output sequence. This approach is analogous to subspace identification for linear stochastic models. A third problem is the estimation problem for hidden Markov models. We show that it suffices for several types of estimation problems to have a solution to the quasi realization problem instead of a solution to the positive realization problem. The techniques are applied to the detection of motifs in DNA sequences.
Concerning the second objective, we consider two modifications to the nonnegative matrix factorization: the structured nonnegative matrix factorization and the nonnegative matrix factorization without nonnegativity constraints on the factors. It turns out that these factorizations are applicable in engineering applications, apart from the hidden Markov research. The structured nonnegative matrix factorization is applied to the clustering of data points based on their distance matrix. The nonnegative matrix factorization problem without nonnegativity constraints on the factors is applied to the modeling of a database containing human faces.

Promotors:
• Prof.dr.ir. B. De Moor, promotor
• Prof.dr.ir. J.C. Willems, copromotor

Tue 29 - Tue 29 Apr-08 OPTEC Seminar - Matthias Gerdts
Auditorium A, ESAT
4:30 pm-5:30 pm
Direct and indirect methods for optimal control problems and applications in engineering
Matthias Gerdts, School of Mathematics, University of Birmingham

Abstract:

The talk discusses numerical methods for optimal control problems with ordinary differential equations and shows how they can be extended to mixed-integer optimal control problems and real-time optimal control problems.

The first part of the talk is concerned with two different approaches for the numerical solution of optimal control problems: the direct discretization approach and a semismooth Newton method. The direct discretization approach transforms the optimal control problem into a nonlinear program which is solved by an SQP method and turns out to be very powerful in practice. The semismooth Newton method is a variant of Newton's method for nonlinear equations and aims at satisfying the first order necessary optimality conditions of the optimal control problem using a nonlinear complementarity function. It possesses good convergence properties and turns out to be particularly well-suited for problems with mixed control-state constraints.

The second part of the talk briefly discusses extensions towards real-time optimization using a sensitivity analysis and mixed-integer optimal control problems using a suitable time transformation.

Finally numerical results for selected applications from aerospace engineering, PDE control and vehicle simulation will be presented.

The organisers gratefully acknowledge the financial support of ICCOS
Some movies shown during the talk are available at: http://web.mat.bham.ac.uk/M.Gerdts/movies.htm

Tue 22 - Tue 22 Apr-08 KU Leuven Workshop on Robust MPC
ESAT - 00.92
9:00 am-5:00 pm
Invited Speakers: M. Cannon, E. Kerrigan and P. Goulart

Programme (.pdf):

The workshop is organized by Carlos Dorea and Moritz Diehl, and co-sponsored by OPTEC as well as ICCoS (Identification and Control of Complex Systems), a Scientific Research Network of the Research Foundation - Flanders (FWO-Vlaanderen).

If you want to participate, send an e-mail to Carlos Eduardo Trabuco Dorea Carlos.Dorea@esat.kuleuven.be with subject "ROBUST MPC WORKSHOP" before April 16.

Mon 21 - Mon 21 Apr-08 Tutorial Course on Robust MPC
ESAT 00.92
9:00 am-5:00 pm
Speakers: Mark Cannon, Moritz Diehl,
Eric Kerrigan and Paul Goulart

Agenda
• 9:00 Welcome and Introduction
• 9:00-13:00   Deterministic MPC
•            9:00 - 10:00 MPC Problem Formulation - Mark
•            10:00 - 10:30 Coffee Break
•            10:30 - 11:30 Stability Analysis and Closed Loop Properties - Mark
•            11:45 - 12.45 Dynamic Programming - Moritz
• 13:00-14:00  Sandwich Lunch
• 14:00-17:00  Numerics and Robust MPC
•            14:00 - 15:00 Numerical Implementation - Paul
•            15:00 - 15:30 Coffee Break
•            15:30 - 16:30 Integral Action and Constant Disturbances - Eric
• 16:30 - 17:00 Discussion
Tutorial Slides

In the afternoon 14-16:30 we might also be in the PC-Pool 00.91 at ESAT.

If you want to participate, send an email to Ida Tassens Ida.Tassens@esat.kuleuven.be with subject "ROBUST MPC TUTORIAL" before April 16.

Tue 15 - Tue 15 Apr-08 Doctoral Presentation - Tom Van Herpe
Promotiezaal Universiteitshal, Naamsestraat 22, 3000 Leuven
5:00 pm
Blood glucose control in critically ill patients:
Design of assessment procedures and a control system

Tom Van Herpe (K.U. Leuven, ESAT-SCD)

Abstract
Critically ill patients, typically admitted to the Intensive Care Unit (ICU), show hyperglycemia and insulin resistance associated with adverse outcomes. It has been demonstrated that strict blood glucose control (between 80 and 110 mg/dl) results in an important reduction in mortality and morbidity. Current therapy requires a manual and rigorous administration of insulin and could, therefore, be replaced by a semi- or fully-automatic blood glucose control system. The introduction of this system could potentially lead to tighter glycemic control and to a decrease of hypoglycemic events and workload of the medical staff.

Three objectives are set in this thesis. The first objective is the design of a procedure to evaluate the reliability of glucose sensor devices (comprising both time-discrete and near-continuous sensors) with regard to a gold standard blood glucose sensor. The quality of blood glucose control depends on the reliability (accuracy) of the measurements. Current methods to assess this reliability level may mislead evaluations and/or lack statistical evidence, however. In this thesis, the GLYCENSIT procedure is developed to assess the performance of a test sensor device with respect to a reference sensor device. We present a method that can be tuned according to the clinician’s preferences regarding significance level, tolerance level, and glycemic range cut-off values. The potential of this new procedure is shown by analysing hypothetical and real-life clinical (ICU) paired glucose data.

The second objective of this dissertation is the design of a procedure to appropriately assess the adequacy of blood glucose control algorithms used in the ICU. Based on clinical expert knowledge, the Glycemic Penalty Index (GPI) is introduced as a measure for the overall glycemic control behaviour in ICU patients as current evaluation measures have important weaknesses that may mislead assessments. Further, the importance of keeping the blood glucose sampling frequency and the duration of algorithm application similar among different patient groups when comparing different insulin titration algorithms, is presented.

The final objective of this thesis is the design of a predictive control system that can potentially be used for (semi-)automatically normalizing the blood glucose in the critically ill. This blood glucose control system comprises a patient model and a controller. Both black-box and grey-box modelling approaches are introduced in this thesis to accurately describe the glucoregulatory system of the critically ill. Only the grey-box modelling approach, expressed in the ICU - minimal model (ICU-MM) and based on physical insight, is found to be valid for use in a clinical predictive control system. Re-estimating the ICU-MM by following an adaptive modelling strategy, allows to capture inter- and intra-patient variability and gives satisfactory forecasting results. Finally, a Model based Predictive Controller that optimizes the insulin infusion rate based on the ICU-MM, is designed. A simulation study shows that the results of the developed control system are satisfactory both in terms of control behaviour (reference tracking and the suppression of unknown disturbance factors) and clinical acceptability.

Promotors

Prof. dr. ir. B. De Moor
Prof. dr. G. Van den Berghe

Mon 14 - Mon 14 Apr-08 IAP VI/4, DYSCO Workshop
K.U.Leuven University Hall, Naamsestraat 22, 3000 Leuven
IAP VI/4, DYSCO Workshop
Organized by K.U.Leuven, (ESAT/SCD)

at
K.U.Leuven University Hall, Naamsestraat 22, 3000 Leuven

(Rooms : Promotiezaal and Jubileumzaal)

More details : http://www.inma.ucl.ac.be/DYSCO/StudyDays/2008/April14.htm

Wed 9 - Wed 9 Apr-08 Simon Stevin Lecture on Optimization in Engineering - Julio Banga
Auditorium of the Arenberg Castle
5:00 pm-6:00 pm
6th Simon Stevin Lecture on Optimization in Engineering

"Optimization in computational systems biology" .pdf

Julio R. Banga, IIM-CSIC,
Spanish Council for Scientific Research
Vigo, Spain
Email: julio@iim.csic.es
Web : http://www.iim.csic.es/~julio/research.html

Abstract

Systems biology aims to discover how function arises from dynamic interactions in living systems. Its ultimate objective should be to explain how the components within a cell interact dynamically, and then how cells interact, resulting in the observed structure, organization and function. Systems biology offers unique opportunities for applying systems engineering methods, which are based on mathematical models. Correspondingly, new types of biological problems are motivating new research in theoretical systems engineering.

Optimization aims to make a system or design as effective or functional as possible. Mathematical optimization methods are widely used in engineering, economics and science. This talk will be focused on applications of mathematical optimization in computational systems biology. Examples are given illustrating the use of optimization methods in topics including optimal model building, optimality in biochemical metabolic networks, optimization of metabolic engineering and synthetic biology. Finally, several perspectives for future research are outlined.

Bibliographical Information:

Julio R. Banga obtained a M.Sc. in Industrial Chemistry from the University of Santiago de Compostela (Spain) in 1988, and a Ph.D. in Chemical Engineering from the same University in 1991. During 1992, he was a postdoc at the University of California, Davis (USA), and then spent three years as Assist. Prof. of Chemical Engineering at the University of Vigo, Spain. During those years, he also spent periods as visiting researcher at the University of Pennsylvania and at the M.I.T. (USA). Since 1996, he is a tenured Scientific Researcher at the Process Engineering Group, IIM-CSIC (Spanish Council for Scientific Research) in Vigo, Spain. His main research topic is the application of optimization methods, with emphasis on global optimization, to problems arising from the domain of nonlinear dynamic processes, with applications targeting the areas of bioprocess engineering and systems biology. He is the author of more than 100 archival publications. Currently he is a member of the Editorial Board of BMC Systems Biology.

The "Simon Stevin Lecture Series on Optimization in Engineering" is set up in order to promote optimization in engineering. For this aim, every quarter of the year an outstanding international scholar is invited to report on latest progress in the development of optimization algorithms and their applications in engineering.
Simon Stevin (1548-1620) was a Flemish mathematician and engineer. Among other, he helped to advance the use of decimal fractions, was the first to explain the tides by the attraction of the moon, and discovered the hydrostatic paradox. He made numerous inventions, among them a wind propelled carriage with sails, the "land yacht", which once impressed Prince Maurice of Orange as it moved faster than horses, in around 1600 on the beach between Scheveningen and Petten. Simon Stevin was fond of promoting the use of science in daily life and in craftmanship, and translated various mathematical terms into dutch. Among other, he introduced the dutch word for mathematics, "wiskunde".

Directly after this spring's Simon Stevin Lecture, a little reception will be given at 18:00 in the salons of Arenberg Castle, to which all attendants of the lecture are most warmly welcome!

***** REGISTRATION ENCOURAGED *****
Please send an e-mail with the subject "STEVIN" to Ida.Tassens@esat.kuleuven.be if you intend to participate in the event. No obligation, just to help us getting an idea how many people plan to come.

This Stevin lecture is co-sponsored by ICCoS (Identification and Control of Complex Systems), a Scientific Research Network of the Research Foundation - Flanders (FWO-Vlaanderen).

Wed 9 - Wed 9 Apr-08 OPTEC Mini Scientific Advisory Board Meeting
Room GEHU00.03
9:30 am-1:00 pm
Wednesday April 9 2008 , 9:30-13:00Room GEHU00.03  in "Geel Huis" Arenberg Kasteelpark 11, left behind ESAT

The aim of the mini Scientific Advisory Board (SAB) Meeting is to present to Julio Banga and Yurii Nesterov what research progress has been achieved in OPTEC in the last year. It will be based only on "poster sessions light" where the the people from each workpackage come to the meeting room for 30 minutes duration (everyone is of course also invited to listen to the other workpackages if she/he likes).

#### Agenda

• 9:30 Introduction Joos Vandewalle and Moritz Diehl
•   10:00 WP 3 - Control System Design: Ilse Smets. Posters by  Diederik Verscheure, Bart Saerens, Boris Houska, Lieboud Van den Broeck, Goele Pipeleers.
•    10:30 WP 4 - Design optimization: Jan Swevers. Posters by Trent McConaghy, Filip Logist, Bram Demeulenaere
•   11:00 WP 1 - Conceptual and theoretical challenges: M. Diehl. Posters by Julian Bonilla, Mauricio Agudelo, Ion Necoara, Samuel Xavier-de-Souza, Michel Baes
•   11:30 WP 2 - System identification and approximation: J.  Suykens. Posters by Marcelo Espinoza, Marco Signoretto, Tillmann Falck, Vanya Van Belle, Fabian Ojeda, Peter Karsmakers, Friedl de Groote
•    12:00 WP 5 - Numerical algorithms and software: S. Vandewalle. Posters by Bart Vandereycken, Iurie Caraus, Joris Vanbiervliet, Carlos Alzate, Toon van Waterschoot
•    12:30 Findings of mini-SAB
•    13:00 end

Fri 21 - Fri 21 Mar-08 K.U. Leuven Seminars on Optimization in Engineering - Stephen Boyd
ESAT 00.62
10:00 am
Relaxed Maximum a Posteriori Fault Identification
Stephen Boyd. Stanford University
joint work with argyrios zymnis and dimitri gorinevski

Abstract

We consider the problem of estimating a pattern of faults, represented as a binary vector, from a set of measurements. The measurements can be noise corrupted real values, or quantized versions of noise corrupted signals, including even 1-bit (sign) measurements. Maximum a posteriori probability (MAP) estimation of the fault pattern leads to a difficult combinatorial optimization problem, so we propose a variation in which an approximate maximum a posteriori probability estimate is found instead, by solving a convex relaxation of the original problem, followed by rounding and simple local optimization.
Our method is extremely efficient, and scales to very large problems, involving thousands (or more) possible faults and measurements. Using synthetic examples, we show that the method performs extremely well, both in identifying the true fault pattern, and in identifying an ambiguity group, i.e., a set of alternate fault patterns that explain the observed measurements almost as well as our estimate.

Slides

Mon 17 - Mon 17 Mar-08 OPTEC Advisory Board Meeting
CS building Celestijnenlaan 200A, ground floor
The yearly OPTEC Scientific Advisory Board (SAB) Meeting will be held at Computer Science building - Celestijnenlaan 200A, ground floor.

Agenda
• 10:30-12:00  Adminstrative Overview of OPTEC (room 00.147, who likes may join)
• 12:00-13:00  Sandwich Lunch in Hall (for all participants, reserve with Ida Tassens with subject SAB-LUNCH before Wed March 12)
• Plenary Poster Presentations (Auditorium)
• 14:15-15:10  Poster Session No. 1 + coffee, GROUP PICTURE
• Plenary Poster Presentations (Auditorium)
• 16:00-17:00   Poster Session 2 + coffee
• 17:00-17:30   SAB internal discussion (room 00.147)
• 17:30-18:00   Communication of SAB Findings to OPTEC Professors and Postdocs (room 00.147)

Wed 12 - Fri 14 Mar-08 Modeling and identification of distributed parameter systems for cell population dynamics
Celestijnenlaan 200 A, B-3001 Heverlee, room 00.147
The aim of the workshop is to create interaction between applied mathematicians, mathematical biologists and immunologists.

The topics include:

•   modeling cell population dynamics (e.g. age-structured models),
•   parameter identification of distributed parameter systems (partial differential equations),
•   flow cytometry based cell turnover experimental studies.

This workshop will also present the results of the project 'Modeling and identification of distributed parameter systems arising in immunology' (Flemish-Russian bilateral scientific cooperation).

Wed 12 - Wed 12 Mar-08 K.U. Leuven Seminars on Optimization in Engineering - H. T. Banks
Dept. of Computer Science, Celestijnenlaan 200A00.144
11:00 am
Inverse Problems: Mathematical and Statistical Methodology
H.T. Banks
Director of Center for Research in Scientific Computation, N.C. State University, Raleigh, USA

Abstract
In this presentation I would give an overview of ordinary least squares (OLS), generalized least squares(GLS) and maximum likelihood estimation(MLE) methods and their use in practical inverse problems.

Sponsored by ICCoS (Identification and Control of Complex Systems).

Tue 11 - Tue 11 Mar-08 K.U. Leuven Seminars on Optimization in Engineering - Lieboud Van den Broeck/Bart Saerens
Mech. Eng. Dept. Rode Zaal, Celestijnenlaan 300B
11:00 am-12:00 pm
Overview of MPC-course at ETH
Lieboud Van den Broeck/Bart Saerens, K.U. Leuven

In this talk  a short overview of the "MPC course" followed at the ETH
Zurich will be given.

The content of the talk includes:
• Introduction to MPC
• Linear MPC (stability, different norms, .)
• Explicit approach
• Hybrid systems
• Software developed at ETHZ
Slides

Thu 6 - Thu 6 Mar-08 K.U. Leuven Seminars on Optimization in Engineering - Ion Necoara
ESAT 00.62
4:00 pm-5:00 pm
Application of a smoothing technique to decomposition in convex optimization
Ion Necoara (K.U. Leuven  ESAT-SCD)

In this talk a  new dual decomposition method  for optimizing a sum of  convex objective functions but with coupling  constraints  corresponding to multiple agents  will be presented. In this method we define a smooth Lagrangian, by using a smoothing technique recently developed  by Nesterov, which preserves separability of the problem. With this  approach we propose a new decomposition method,  for which efficiency estimates  are derived and which improves the bounds on the number of iterations of the classical dual gradient scheme by an order of magnitude. The method involves every agent optimizing an
objective function that is the sum of his own objective function and a smoothing term while the coordination is performed via the Lagrange multipliers. Application of the new method for solving distributed model predictive control problems with coupling dynamics  and decoupled costs is also illustrated.

Fri 29 - Fri 29 Feb-08 Gene Golub Commemoration Event
tba
9:00 am-5:00 pm
This event is part of the Gene Golub Around the World Day.

Thu 28 - Thu 28 Feb-08 K.U.Leuven Seminars on Optimization in Engineering - Oliver Stein
ESAT 00.62
5:00 pm-6:00 pm
"On Gemstones and Maneuverability Problems: Applications of Modern Design Centering Techniques"  Oliver Stein  (University of Karlsruhe (TH) )

Design centering problems deal with the inscription of a variable body into a fixed container so that, for example, the volume of the body is maximized. Applications include
• the maximal inscription of a gemstone into a rough stone to minimize the wasted material,
•  the computation of lower bounds for the volume of complicated container sets by inscription of balls, like in the manoeuvrability problem in robotics, and
•  the determination of “innermost points” of sets in order to stay away from their boundaries, like in quality control of production processes.
A numerical solution method for design centering problems with irregular geometrical shapes, in particular in the absence of convexity, is presented. The focus of this method is to produce feasible iterates, that is, each iterate corresponds to a body which is a guaranteed subset of the container. To enforce feasibility, our method constructs certain convex relaxations with ideas from the alpha-BB method of global optimization. Numerical examples illustrate the performance of the method.

Directly after the seminar, the attendants are warmly invited to an "optimization coffee" for discussion.

Slides

Mon 25 - Mon 25 Feb-08 SISTA Seminar - Mario Figueiredo
ESAT 00.62
4:00 pm-5:00 pm

"Semi-supervised Classification and Clustering"

Mario A. T. Figueiredo
Instituto de Telecomunicações
Instituto Superior Técnico
Lisboa, PORTUGAL

Recently, there has been considerable interest in non-standard learning scenarios, namely in the so-called semi-supervised learning problems. Most formulations of semi-supervised learning see the problem from one of two (symmetrical) perspectives: supervised learning (namely, classification) with some (maybe many) missing labels; unsupervised learning (namely, clustering) with additional information. In this talk, I will review recent work in these two areas, with special emphasis on our own work. For semi-supervised learning of classifiers, I will describe an approach which is able to incorporate unlabelled data as a regularizer for a (maybe kernel) classifier. Unlike previous approaches, the method is non-transductive, thus computationally inexpensive to use on future data. For semi-supervised clustering, I will present a new method, which is able to incorporate pairwise prior information in a computationally efficient way, and which particularly well suited for image segmentation.

Thu 14 - Thu 14 Feb-08 K.U. Leuven Seminars on Optimization in Engineering - Mark Cannon
ESAT 00.62
5:00 pm-6:00 pm
"Stochastic Model Predictive Control"
Mark Cannon, University of Oxford

Stochastic Model Predictive Control is emerging as an area of research of significant theoretical interest which addresses problems of practical importance. Model Predictive Control (MPC) is recognized as the methodology of choice for optimizing closed loop performance in the presence of constraints, and it is unique in providing computationally tractable optimal control laws by solving constrained receding horizon control problems online. However, most real life applications are not only subject to constraints but also involve multiplicative and/or additive stochastic uncertainty. Earlier work tended to ignore information on the distribution of model uncertainty, and as a result addressed control problems suboptimally using robust MPC strategies that employ only information on bounds on the uncertainty. This talk will discuss strategies for handling probabilistic constraints for the case of uncertain linear systems, and will present methods of optimizing performance subject to this type of constraint, and also demonstrate that constraint satisfaction is achieved in closed loop operation. The solution of a mixed objective control problem involving coupled algebraic Riccati equations (CARE) will be described, and the convergence properties of the associated constrained receding horizon control problem presented. The results are illustrated by a design study considering control of a wind turbine in order to maximize power capture subject to constraints on fatigue damage.

Wed 30 - Wed 30 Jan-08 OPTEC Miniworkshop on "Optimal Control for Solar Thermal Power Plants"
ESAT 01.60
9:00 am-12:00 pm
Program:

Press Release on newly founded
Virtual Institute of Central Receiver Solar Power Plants

Wed 23 - Wed 23 Jan-08 K.U. Leuven Seminars on Optimization in Engineering - Paul Williams
ESAT 01.60
2:00 pm-3:00 pm
"Dynamics and Control of Tethered Systems with Applications to Kite Power"
Dr. Paul Williams (T.U. Delft / Australia)

This presentation will begin by giving an overview of tether dynamic modelling. Various applications of the model will be shown in diverse environments such as air, sea, and space. Simplifications that can be made for efficient control development will be described, which are validated through simulation and flight on the Young Engineer’s Satellite 2. Methods for trajectory optimization of general dynamical systems using pseudospectral and quadrature methods will be presented and typical results will be given for a range of applications. The above modelling and control techniques are fused to enable development of kite trajectories for power generation. Simplified models for kite control design will also be presented.

Fri 18 - Fri 18 Jan-08 K.U. Leuven Seminars on Optimization in Engineering - Tamas Keviczky
Esat 00.62
11:00 am-12:00 pm

"Distributed Constrained Optimal Control of Large-Scale Interconnected Systems"
Tamas Keviczky (T.U.Delft)

Large-scale engineering systems composed of interconnected components can be found in a broad spectrum of applications ranging from networks of independently actuated vehicles in robotics and formation flight to industrial process control and civil engineering. The presentation will focus on distributed control design methods for these systems using constrained optimal control techniques. Interest in such problems is usually motivated by the necessity of avoiding centralized design when this becomes computationally prohibitive or would require unrealistic expectations regarding information exchange. In particular, a distributed predictive control scheme will be described with a discussion on stability and feasibility related questions. Recent results in distributed design for a special class of large-scale systems using LQR controllers will also be summarized.

Applicability of the proposed framework will be demonstrated using the formation control problem of multiple Unmanned Air Vehicles (UAVs) as a motivating example. This particular application problem has a wide range of envisioned applications including distributed sensing and monitoring, which appear to be the most promising ones. The challenge in UAV formation flight is to formulate simple local problems and design individual controllers which would enable the UAVs to meet certain maneuvering challenges while maintaining relative positions and safe distances between each vehicle. This objective is achieved by developing distributed control laws for cooperative multi-vehicle groups using constrained optimal control theory and their demonstration on high-fidelity models of an actual unmanned hovering vehicle platform.

Slides

Wed 19 - Wed 19 Dec-07 Simon Stevin Lecture on Optimization in Engineering - David Mayne, Guido Vanden Berghe
Auditorium of the Arenberg Castle
4:00 pm-6:45 pm

Fifth Simon Stevin Lecture on Optimization in Engineering

Program:

16:00 Simon Stevin Lecture on Optimization in Engineering by David Mayne
17:00 Short Coffee Break
17:15 OPTEC Christmas lecture by Guido Vanden Berghe
18:15 Reception in the salons of the Arenberg castle

Flyer ,   Poster

Simon Stevin Lecture on Optimization in Engineering

"Optimization: highways and byways"
David Mayne (Imperial College London)

Abstract:
I commenced my research in optimal control and optimization based design in 1959 so I have seen great changes in our subject over nearly 50 years.  In this talk, I outline some areas of research that I have been engaged in, some of which developed into major activities (the highways) and some of which remained dormant (byways). Both are interesting since even the byways may suggest fruitful areas for future research. The topics that I discuss include algorithms for optimal control (differential dynamic programming and strong variation algorithms), Monte Carlo procedures for stochastic control problems and filtering (particle filters), design problems involving infinite dimensional constraints such as the tuning problem in circuit design, and model predictive control. My aim is to present the relatively simple concepts that motivated research in these areas.

Bibliographical Information:
David Mayne received the B.Sc. and M.Sc degrees in Engineering from the University of the Witwatersrand, the Ph.D. and D.Sc degrees from the University of London, and the degree of Doctor of Technology, honoris causa, from the University of Lund, Sweden. He has held posts at the University of the Witwatersrand, the British Thomson Houston Company, University of California, Davis and Imperial College London where he is now Senior Research Fellow. He has been awarded Fellowships from the Royal Society, Royal Academy of Engineering, IEEE, IFAC, IEE and Imperial college. He has held numerous visiting professorship appointments.

David Mayne left traces in nearly all areas of control. His research interests include optimization, optimization based design, nonlinear control, model predictive control, and adaptive control.

Slides

OPTEC Christmas lecture

"Simon Stevin (1548-1620) Mathematician, physicist, engineer . . . , Uomo universale"
Guido Vanden Berghe (Universiteit Gent)
Guido.VandenBerghe@UGent.be

Abstract:
In this talk we shall give in first instance attention to the family and the life of Simon Stevin. Born in Bruges his grandparents were original wealthy inhabitants of Ypres and Veurne. In the second place we shall present a comprehensive picture of the activities and the creative heritage of Simon Stevin, who made outstanding contributions to various fields of science in particular, physics and mathematics and many more. Among the striking spectrum of his ingenious achievements, it is worth emphasizing, that Simon Stevin is rightly considered as the father of the system of decimal fractions as it is in use today. Stevin also urged the universal use of decimal fractions along with standardization in coinage, measures and weights. This was a most visionary proposal.
Stevin was the first since Archimedes to make a significant new contribution to statics and hydrostatics. His activities as an engineer will be discussed; in particular the construction of fortifications,windmills and the famous sailing chariot will be illustrated. He truly was an "uomo universale".

References
(1) J.T. Devreese en Guido Vanden Berghe, Wonder en is gheen wonder , De geniale wereld van Simon Stevin. 1548-1620, Davidsfonds, Leuven, 2003.
(2) J.T. Devreese and Guido Vanden Berghe, Magic is No Magic: The Wonderful World of Simon Stevin , WITpress Southampton (UK), 2007 .

Slides

The "Simon Stevin Lecture Series on Optimization in Engineering" is set up in order to promote optimization in engineering. For this aim, every quarter of the year an outstanding international scholar is invited to report on latest progress in the development of optimization algorithms and their applications in engineering.
Simon Stevin (1548-1620) was a Flemish mathematician and engineer. Among other, he helped to advance the use of decimal fractions, was the first to explain the tides by the attraction of the moon, and discovered the hydrostatic paradox. He made numerous inventions, among them a wind propelled carriage with sails, the "land yacht", which once impressed Prince Maurice of Orange as it moved faster than horses, in around 1600 on the beach between Scheveningen and Petten. Simon Stevin was fond of promoting the use of science in daily life and in craftmanship, and translated various mathematical terms into dutch. Among other, he introduced the dutch word for mathematics, "wiskunde".

Directly after this winter's Simon Stevin Lecture, a little reception will be given at 18:45 in the salons of Arenberg Castle, to which all attendants of the lecture are most warmly welcome!

***** REGISTRATION ENCOURAGED *****
Please send an e-mail with the subject "STEVIN" to Ida.Tassens@esat.kuleuven.be if you intend to participate in the event. No obligation, just to help us getting an idea how many people plan to come.

This Stevin lecture is co-sponsored by ICCoS (Identification and Control of Complex Systems), a Scientific Research Network of the Research Foundation - Flanders (FWO-Vlaanderen).

Tue 11 - Tue 18 Dec-07 Convex Optimization 3 Day Mini Course - Michel Baes
ESAT 00.91
10:35 am-12:35 pm

Convex Optimization 3 Day Mini Course
by Michel Baes

Date:  Dec 10, 12 and 18,
Time:  10:35-12:35,
Place: ESAT 00.91 (PC Pool)

In this two session lecture with exercises, we deal exclusively with convex optimization  problems. We fist show how every convex optimization problem can be put into a so-called conic format, which enables us to use the powerful tools of duality theory to solve it. We show how several practical optimization problems can be formulated as instances of conic programming, and how we can solve most of them extremely efficiently using the software Sedumi. We will also describe briefly the main ideas of the primal-dual algorithm used in Sedumi. Several examples will be set up by the participants.
The exercise-lectures are part of the master course on "Numerical
Optimization" by Moritz Diehl but open to some extra participants "Convex conic programming"

"Duality Theory" (11/12/2007)

Duality originates from the very simple idea that one can obtain
combining its constraints. Yet, this simple idea has an impressive
number of consequences, in a theoretical level with a.o. the development
of the Lagrangean framework, and the foundations of an elegant
differentiability theory, in a practical level with a.o. the economic
interpretation of dual variables and the development of extremely
efficient primal-dual algorithm (this last topic will be sketched in the
next lecture).

"Convex conic programming" (12/12/2007)

In this lecture, we deal exclusively with convex optimization problems.
We first show how every convex optimization problem can be put in a
so-called conic format, which enables us to use the powerful tools of
duality theory to solve it. We show how several practical optimization
problems can be formulated as instances of conic programming, and how we
can solve most of them extremely efficiently using the software Sedumi.
We will also describe briefly the main ideas of the primal-dual
algorithm used in Sedumi.

Fri 7 - Fri 7 Dec-07 SISTA Seminar - Dennis Bernstein
ESAT 01.60
10:00 am-11:00 am

Mysteries and Conundra in the
Use of Physical Dimensions

Dennis Bernstein
(Department of Aerospace Engineering, University of Michigan, Ann Arbor, Michigan, USA)

Physical dimensions are the glue that binds mathematics to physical reality. We use these quantities so routinely that we scarcely give them any thought. Nevertheless, there are a few aspects of physical dimensions that require careful attention and that are not fully settled. In this talk I will first discuss basic concepts relating to physical dimensions, including a proof of the cornerstone (and only) result in the field, namely, the Buckingham Pi theorem. Next, I will discuss various philosophical and technical subtleties that arise in the use of dimensions. For example:

Which approach is better: a) nondimensionalizing all physical quantities, b) working with dimensioned quantities directly, or c) ignoring dimensions completely (as in control theory)?

Why do dimensions matter when all real measurements are in volts?

Are physical dimensions relevant when we run a Matlab code with billions of operations on numbers?

If radians are dimensionless, why don’t we just ignore them completely?

Are all dimensionless quantities equivalent? Is there a difference between radian and Reynolds number?

Don’t energy and moment have the same dimensions, namely, length times distance? If so, what does it mean for a moment to do work?

Can a matrix with dimensioned entries have a determinant? Can it have eigenvalues and eigenvectors? If so, what are their dimensions?

Is it legal for control theorists to write V = xdot^2 + x^2?

Based on joint work with Harish Palanthandalam-Madapusi and Ravinder Venugopal.

Fri 7 - Fri 7 Dec-07 Doctoral Presentation - Steven Gillijns
Auditorium of the Arenberg Castle
5:00 pm
"Kalman filtering techniques for system inversion and data assimilation"
Steven Gillijns (K.U. Leuven, ESAT-SCD)

Abstract
Since its introduction in 1960, the Kalman filter has gained increasing popularity.
It has become the standard technique for estimating the present state
of a dynamical system based on a numerical model of that system and a set
of observations. This thesis contributes to the popularity of the Kalman filter
by addressing the problems of system inversion and data assimilation from the
viewpoint of Kalman filtering.

In applications such as fault detection and cryptography, the dynamical
system is subject to inputs that are unknown, but yet are of major importance.
The problem of estimating the inputs of a dynamical system from observations
of that system’s outputs, has been termed system inversion. In the first part of
this thesis, a new inversion procedure based on joint input-state estimation is
developed. Conditions are derived under which the poles of the estimator can
be assigned and the speed of convergence can thus be tuned. In case of noise,
it is shown that the poles can be placed so that, in analogy to the Kalman filter,
the estimates of the system state and the system input are optimal in a
least-squares sense. Several computational and numerical issues such as
reduced order estimation and square-root estimation are addressed. The
inversion procedure is employed in four applications.

Due to its high computational cost and its immense storage requirements,
the Kalman filter is not directly applicable with the large-scale numerical models
that are usually employed in environmental problems such as weather
prediction. The challenging problem of assimilating observations in such
complex numerical models has been termed data assimilation. In the second
part of this thesis, data assimilation techniques are developed for nowcasting a
space weather event that emulates the topology and the dynamics of the bow
shock that is formed when the supersonic solar wind encounters the Earth. A
suboptimal Kalman filter is developed that is adapted to the data-sparse
environment of space weather. Simulation results on a large-scale model show
that the estimates produced by the new suboptimal filter outperform a data-free
simulation, even if only a few observations are available.

Promotor
Prof.dr.ir. B. De Moor

Mon 3 - Mon 3 Dec-07 SISTA Seminar - Renato Markovinovic
Room 00.62
3:30 pm-4:30 pm
Employing Model Reduction for Efficient Solution of Reservoir Engineering Problems

Renato Markovinovic, T.U. Delft, NL

The talk gives an overview of the main research activity of the speaker in the last few years:
the employment of (generally) projection-based model order reduction (MOR) methods for solving multiphase porous media flow problems.
The starting point of modeling porous media fluid flow for purposes of simulation, optimization and/or estimation are, generally: a) partial differential equations describing the conservation of mass of the different fluid-phases involved, b) an empirically derived momentum equation relating the fluid-flow rate and the pressure (or,more generally, potential) gradient, and c) 'closure' relations, necessary to obtained a well-defined system of equations. Reservoir flow problems (single and multi-phase, multicomponent porous media flow problems) appear to be a real challenge for any MOR strategy, due to a large scale of different phenomena, both physical and numerical (severe nonlinearities, possibly shocks, moving fluid-front interfaces, coupled physics: convection-diffusion, a mix of Parabolic, Hyperbolic and Eliptic PDEs, highly heterogeneous parameter properties, very large models (O(10^7) is not an exception), etc., etc.). For single-phase flow, which is often possible to model by linear equations, the MOR methods considered range from transfer function moment matching ones (both explicit matching, as e.g. AWE, and implicit matching by Krylov subspaces), via Modal Truncation, till the more system-theoretic based techniques of (approximate) balanced truncation. A new development for this type of large-scale systems is the employment of "non-vectorizing POD-like" methods inspired by the recently developed algorithms of high-dimensional generalizations of PCA and alike (HOOI, GLRAM, 2DPCA, etc.). For multi-phase flow systems, due to their physical and mathematical complexities, the investigation has so far been restricted to the standard POD method and its variations (e.g., Sobolev norm based, Adaptive POD as predictor for the high-dimensionale iterative solutions, etc.). Finally, subspace identification as a tool to obtain low-dimensional single-phase flow models from input-output (simulation!) data has also been investigated.

Mon 26 - Fri 7 Dec-07 Graduate school in systems, control and networks course "Numerical methods for nonlinear optimal control problems" - Moritz Diehl
Esat room 00.62 and 02.58
9:15 am-12:15 pm

This is a 5 day course within the Graduate school in Systems, Optimization, Control and Networks (SOCN)

NUMERICAL METHODS FOR NONLINEAR OPTIMAL CONTROL PROBLEMS

Dates : November 26, 29, and December 03, 06, 07, 2007
Schedule : 9h15 - 12 h45

Announcement: SOCN website

Lecturer : M. DIEHL

The course develops numerical solution strategies for optimization problems with underlying differential equation models. After a brief overview over sequential and simultaneous approaches to optimal control, we focus on the latter class, which solve both the model equations and the optimality conditions in a "one-shot' approach. Here, exploitation of sparsity in the linear solvers is of crucial importance. Special topics include the treatment of distributed inequalities by interior point methods, periodic problems. Finally, we discuss real world applications from chemical and mechanical engineering.

Contents
1. Optimal Control: Introduction and Overview
2. Dynamic Programming and Indirect Approaches (Pontryagin)
3. Direct Transcription Methods (ODE Sensititivity, Single Shooting)
4. Direct Collocation, Treatment of Sparsity
5. Direct Multiple Shooting
6. Applications and Extensions: Multi-Stage Processes, Periodicity

Prerequisites
Participants are assumed to have a knowledge of linear algebra and calculus. Basic knowledge in optimisation and numerical simulation of dynamic systems is advantageous.

The course will take place at Esat in room 00.62, except on  26/11 it will be in room 02.58.

Thu 22 - Thu 22 Nov-07 DYSCO (Dynamical Systems, Control, and Optimization) Study Day
Brussels
9:30 am-4:30 pm

Nov 22, 9:30-16:30, Brussels: DYSCO (Dynamical Systems, Control, and Optimization) Study Day :
http://www.inma.ucl.ac.be/DYSCO/StudyDays/2007/November22.htm

Thu 22 - Thu 22 Nov-07 Seminar by Philippe Toint
Namur
2:00 pm-4:00 pm

"Adaptive cubic overestimation for unconstrained optimization"
Philippe Toint

http://www.fundp.ac.be/facultes/sciences/departements/mathematique/agenda/evenement.2007-11-06.8872619258

Tue 20 - Tue 20 Nov-07 K.U. Leuven Seminars on Optimization in Engineering - Samira Roshany
Esat 02.54
6:30 pm-7:00 pm
"Flight parameter estimation in frequency domain and comparison with time domain"
Samira Roshany (Iran)

In this study, at first parameter estimation has been introduced as a science identifying all systems around us. And a perspective of the possible solutions and algorithms in parameter estimation has been given. The method of parameter estimation in frequency domain has been surveyed as suitable estimation approach for linear systems. And its results have been compared to the results of estimation in time domain method. In this project we used the Equation Error method to estimate the unknown parameters and the mathematical model was state space model. To minimize the error that produced by disturbance, we use the Least Square cost function and optimal control methods. The frequency domain method was the FTR approach, which uses the Fourier Transform Regression to invert the parameters from time domain to frequency domain.
In these study 18 parameters of stability and control derivatives for Cessna172 aircraft, have been estimated from which 11 were about lateral motion, 5 about Short period mode in longitudinal motion and 2 about Phoguid mode in longitudinal motion. By comparing parameter estimation in two domains relative to nominal value it has been identified that the amount of error in frequency domain approach were much more than the error in time domain method. The amount of the error was 15.3 percent for lateral motion derivatives, 9.6 percent for longitudinal motion derivatives whereas the amount of the error in time domain method was 6.4 percent for lateral motion derivatives and 2 percent for longitudinal motion derivatives. But as we don t have the nominal value in parameter estimation, we have preferred to use equation error variance or mean square error (MSE), instead of comparing two methods based on relative error average. In the new method it has been noticed that again variance error in frequency domain is more than that in time domain.
At last the reasons for these errors have been surveyed. And methods to improve the estimating results in frequency domain have been proposed.

Mon 19 - Fri 23 Nov-07 ATHENS course "Dynamic Process and System Optimization" - Moritz Diehl
to be defined
9:00 am-5:00 pm

ATHENS course "Dynamic Process and System Optimization" November 19-23
teachers: Moritz Diehl, Bart Saerens, Dang Doan, Niels Haverbeke

Program and room information can be found on the course website.

Thu 15 - Thu 15 Nov-07 K.U. Leuven Seminars on Optimization in Engineering - Paul Van Dooren
ESAT - Room 00.62
4:00 pm-5:00 pm

"Some graph optimization problems in data mining"
Paul Van Dooren, UCL

Graph-theoretic ideas have become very useful in uderstanding modern large-scale datamining techniques. We show in this talk that ideas from optimization are also quite useful to better understand the numerical behaviour of the corresponding algorithms. We illustrate this claim by looking at two specific graph theoretic problems and their application in datamining.
The first problem is that of reputation systems where the reputation of objects and voters on the web are estimated; the second problem is that of estimating the similarity of nodes of large graphs. These two problems are also illustrated using concrete applications in datamining.

Slides

Fri 9 - Fri 9 Nov-07 K.U. Leuven Seminars on Optimization in Engineering - Bram Demeulenaere
Celestijnenlaan 200A room 00.147
1:00 pm-2:00 pm

"Linear programs for optimal motion generation in linear dynamic systems: numerical issues"
Bram Demeulenaere (K.U.Leuven Mech-PMA)

Abstract:
Since 2005, a linear-programming based optimization framework has been developed at KUL/PMA to design dynamically optimal, polynomial splines for motion generation. While the framework was originally developed for rigid-body mechanical systems, it has recently been extended to flexible systems, described by linear ODEs subject to parametric uncertainty. For the latter class of systems, also so-called input shaping techniques have been considered, whereby it was shown in a 2006-2007 master thesis that input shapers can also be designed based on linear programming.

The present talk gives a short and high-level introduction to both the polynomial spline and input shaper optimization, and subsequently focuses on the main problems encountered so far, which are predominantly numerical in nature. The main idea of the talk is (i) to get some preliminary feedback of specialists in the field of spline optimization, linear algebra/programming and general scientific computing, and (ii) to investigate whether a collaboration concerning this topic can be set up.

Note:
Bram Demeulenaere is a member of OPTEC and an expert on the application of convex optimization techniques to mechanical systems.  He works in particular on counterweight balancing of linkages, spline-based cam design, dynamically compensated cams, and inverse dynamic musculo-skeletal simulation, including muscle physiology.

Slides

Thu 8 - Thu 8 Nov-07 Workshop on Nonlinear State Estimation
Esat, Room 00.62
2:00 pm-8:00 pm

"Workshop on Nonlinear State Estimation"
November 8, 2007
Katholieke Universiteit Leuven

Aim:

State and parameter estimation of nonlinear dynamic systems is an important prerequisite for their control and optimization. It is an integral part in such diverse applications as process monitoring, fault detection, process optimization, and model predictive control.

This OPTEC topical workshop will focus on the question of how to estimate the state of a dynamic system accurately and fast, given online measurements and a possibly nonlinear model. It is intended as a forum that shall bring young researchers from within the OPTEC environment together that work on nonlinear state estimation.
It is opened by a lecture by Johannes P. Schloeder from the IWR in Heidelberg, who reviews Numerical Methods for Parameter and State Estimation in Nonlinear Differential Equation Systems.

Confirmed speakers:

• Johannes Schloeder, IWR University of Heidelberg, Germany
• Friedl DeGroote, K.U.Leuven - PMA
• Tinne De Laet, K.U.Leuven - PMA
• Niels Haverbeke, K.U.Leuven - SCD
• Jeroen Boets, K.U.Leuven - SCD
• Wolfgang Mauntz, University of Dortmund, Germany
• Geert Gins, K.U.Leuven - BIOTEC

Wed 31 - Wed 31 Oct-07 K.U. Leuven Seminars on Optimization in Engineering - Mike Powell
Esat 01.57
4:00 pm-5:00 pm
"Some recent research on minimization without derivatives"
M.J.D. Powell  (University of Cambridge)

The NEWUOA software for unconstrained minimization without derivatives
employs quadratic models of the objective function, F say. Each model
has (n+1)(n+2)/2 parameters, where n is the number of variables, but,
when solving some test problems, the total number of calculations of
F is only of magnitude n. This mystery will be explained. Each new model
is derived from the old model and about 2n+1 interpolation conditions,
the amount of routine computation of each iteration being only O(nn).
Thus several functions of 320 variables have been minimized successfully.
Recent research to be mentioned addresses moves of interpolation points
that are far from the best point so far, the number of interpolation
conditions, and the inclusion of prescribed lower and upper bounds on
the variables.

Mon 29 - Tue 30 Oct-07 Symposium on "The birth of numerical analysis"
Auditorium of the new Computer Science building, Celestijnenlaan 200 A
10:00 am

Conference website

At this conference also Mike Powell will give a talk with the title "The development of algorithms for nonlinear optimization".

Mike Powell is (emeritus) Professor at Cambridge university and one of the most important and influential nonlinear optimizers worldwide.
He left traces in several fields of optimization, among other with the invention of the now widespread SQP algorithm 30 years ago, up to efficient derivative free nonlinear optimization algorithms in recent years.

Fri 26 - Fri 26 Oct-07 K.U. Leuven Seminars on Optimization in Engineering - Mark Eifert
Esat 00.62
5:00 pm-6:00 pm

"Energy Management in Conventional and Hybrid Vehicles as an Optimization Problem"
Mark Eifert

Mechanical power is obtained through a combustion process in an engine, and a part of that power may be converted to electrical power with an electric machine. The electrical power may either be buffered in an energy storage device for future use or used immediately by electrical consumers or, in the case of a full or mild hybrid system, by an electric traction motor. The control of the generation of electrical power, the control of an electrical machine as a motor and the scheduling of electrical consumers are all objectives of an energy management strategy. Such strategies generally have the goals of maintaining enough stored electrical energy to guarantee the ability to start the vehicle and always provide requested electrical power and to lower fuel consumption and emissions. They are developed with system models, whose fidelity and accuracy indirectly determine the approach and complexity of the strategy itself. A number of different approaches are possible when designing an energy management strategy. They include predictive control, which requires a prediction of vehicle states over a time horizon, or game theory, which poses the optimal control problem as a game between the controller and the environment, where neither player knows what the other will do next. In my presentation, I would like to discuss the development of energy management strategies for conventional and hybrid vehicles. In it, I will touch on basic vehicular power supply and hybrid architectures, components and their modeling. I will then discuss the optimization problem and some strategy design approaches.

Slides

Thu 25 - Thu 25 Oct-07 K.U. Leuven Seminars on Optimization in Engineering - Colin Jones
Esat 00.62
4:00 pm-5:00 pm

"Parametric Linear Complementarity Problems in Control"
Colin Jones (ETH, Zurich)

Abstract:
Constrained finite time optimal control problems can be expressed as
mathematical programs parameterized by the current state of the
system: the so-called multi-parametric programs. These problems have
received a great deal of attention in the control community during the
last few years because solving the parametric program is equivalent to
synthesizing the optimal state-feedback controller. For many cases of
interest, the resulting synthesized controllers are simple
piecewise-affine functions, which enables receding horizon control to
be used not only in slowly sampled systems requiring powerful
computers but now also in high-speed embedded applications running at
many kilohertz or megahertz.

In this talk, we introduce the parametric linear complementarity
problem (pLCP), which unifies and generalizes linear and quadratic
programs and bimatrix games. This problem allows the synthesis of
constrained receding horizon controllers for linear systems, based on
the optimization of quadratic or piecewise-linear costs. We also find
that many fundamental algorithms of computational geometry that are of
interest in various areas of constrained linear control can be posed
as an equivalent convex parametric LCP.

We present a new computational method for solving this important class
of problems. The method is shown to be polynomial time in the output
size when the problem is in general position. Furthermore, we describe
how the symbolic technique of lexicographic perturbation can be
applied to simulate general position and thus extend the algorithm to all
convex degenerate LCPs.

Slides

Wed 24 - Wed 24 Oct-07 Simon Stevin Lecture on Optimization in Engineering - Manfred Morari
Auditorium of the Arenberg Castle
5:00 pm-6:00 pm

Fourth Simon Stevin Lecture on Optimization in Engineering

"Control of Hybrid Systems: Theory, Computation and Applications"
Manfred Morari (ETH Zurich)

Abstract:    slides      poster

Theory, computation and applications define the evolution of the field of control. This premise is illustrated with the emerging area of hybrid systems, which can be viewed, loosely speaking, as dynamical systems with switches. Many practical problems can be formulated in the hybrid system framework. Power electronics are hybrid systems by their very nature, systems with hard bounds and/or friction can be described in this manner and problems from other domains, as diverse as driver assistance systems, anesthesia and active vibration control can be put in this form.

I will describe the theoretical basis of some of the tools that have been proposed to synthesize the controllers for hybrid systems. Parametric programming has received a lot of attention in the control literature in the past few years because model predictive controllers (MPC) can be posed in a parametric framework and hence pre-solved offline, resulting in a significant decrease in on-line computation effort. I will describe recent work on parametric linear programming (pLP) from the point of view of the control engineer. I will survey various types of algorithms, and identify a new standard convex hull approach that offers significant potential for approximation of pLPs for the purpose of control. The resulting algorithm, based on the beneath/beyond paradigm, computes low-complexity approximate controllers that guarantee stability and feasibility.

Many industrial applications will serve to highlight the theoretical developments and the extensive software that helps to bring the theory to bear on the practical examples.

joint work with Colin Jones, Miroslav Baric and Melanie Zeilinger.

Bibliographical Information:
Manfred Morari was appointed head of the Automatic Control Laboratory at ETH Zurich in 1994. Before that he was the McCollum-Corcoran Professor of Chemical Engineering and Executive Officer for Control and Dynamical Systems at the California Institute of Technology. He obtained the diploma from ETH Zurich and the Ph.D. from the University of Minnesota, both in chemical engineering. His interests are in hybrid systems and the control of biomedical systems. In recognition of his research contributions, he received numerous awards, among them the Donald P. Eckman Award and the John Ragazzini Award of the Automatic Control Council, the Allan P. Colburn Award and the Professional Progress Award of the AIChE, the Curtis W. McGraw Research Award of the ASEE, Doctor Honoris Causa from Babes-Bolyai University, Fellow of IEEE, the IEEE Control Systems (Technical Field) Award, and was elected to the National Academy of Engineering (U.S.). Professor Morari has held appointments with Exxon and ICI plc and serves on the technical advisory board of several major corporations.

The "Simon Stevin Lecture Series on Optimization in Engineering" is set up in order to promote optimization in engineering. For this aim, every quarter of the year an outstanding international scholar is invited to report on latest progress in the development of optimization algorithms and their applications in engineering.
Simon Stevin (1548-1620) was a Flemish mathematician and engineer. Among other, he helped to advance the use of decimal fractions, was the first to explain the tides by the attraction of the moon, and discovered the hydrostatic paradox. He made numerous inventions, among them a wind propelled carriage with sails, the "land yacht", which once impressed Prince Maurice of Orange as it moved faster than horses, in around 1600 on the beach between Scheveningen and Petten. Simon Stevin was fond of promoting the use of science in daily life and in craftmanship, and translated various mathematical terms into dutch. Among other, he introduced the dutch word for mathematics, "wiskunde".

Directly after this autumn's Simon Stevin Lecture, a little reception will be given at 18:00 in the salons of Arenberg Castle, to which all attendants of the lecture are most warmly welcome!

***** REGISTRATION ENCOURAGED *****
Please send an e-mail with the subject "STEVIN" to Ida.Tassens@esat.kuleuven.be if you intend to participate in the event. No obligation, just to help us getting an idea how many people plan to come.

This Stevin lecture is co-sponsored by ICCoS (Identification and Control of Complex Systems), a Scientific Research Network of the Research Foundation - Flanders (FWO-Vlaanderen).

Tue 16 - Tue 16 Oct-07 K.U. Leuven Seminars on Optimization in Engineering - Peter Ortner
Esat 00.62
5:00 pm-6:00 pm

"MPC for Diesel engine air path control"
Peter Ortner, Roland Bergmann

Diesel engines are nowadays equipped with an exhaust gas recirculation valve (EGR) and a variable geometry turbocharger (VGT) to influence the emissions and the torque output. The engine control problem can be stated as an optimization problem including fuel consumption, torque and emmissons. Due to complications in measuring the emissions in a production car an equivalent problem, known as the air path control problem, will be formulated. In the air path control subsitute quantities are used; the two most used quantities are the fresh air mass flow through the compressor (MAF) and the inlet manifold absolute pressure (MAP). These quantities are easily measured in a production car engine and directly affect the torque, fuel consumption and emissions.

A standard engine control unit (ECU) uses two single input single output (SISO) gain scheduled control loops to control MAP with VGT and MAF with EGR, which again requires a huge amount of tuning. This control strategy ignores that EGR also influences MAP and VGT also influences MAF via the so called cross coupling e.g. ignoring that the air path control system is a coupled MIMO system. Therefore it seems obvious to make use of a MIMO control strategy.

For air path modelling in the literature first principles are commonly used leading to mean value models (MVM) which are usually simplified for control design e.g. locally linearized except in the NMPC case. This motivates to directly use data based linear system identification techniques for the modelling. Due to the nonlinear behaviour of the air path no single linear model will give a satisfying quality which leads e.g multi linear identification techniques, LPV identification, a clustering technique for piecewise affine identification or recursive online estimation.

Experiment results, performed on a dynamical engine test bed, show an immense improvement in terms of MAF and MAP tracking of an explicit MPC controller compared to a standard ECU controller. As an alternative for explicit MPC, which is designed to be real time applicable, an online active set strategy is used for fast solving of the quadratic programs online. This leads to a reduction in computing power and therefore to higher possible complexity e.g. higher control horizons.

Results also claim that for affecting emissions (mainly nitrous oxides and particulate matter) in a positive way it is not enough to change only the control scheme also the setpoints for MAF and MAP need to be modified.

Slides

Tue 16 - Tue 16 Oct-07 K.U. Leuven Seminars on Optimization in Engineering - Lorenzo Fagiano
Esat 00.62
4:15 pm-4:45 pm

"Fast implementation of MPC using Set Membership approximation methodologies"
Lorenzo Fagiano (Politecnico di Torino)

Model Predictive Control (MPC) (see e.g. the survey [1]) is a model based control technique where the control action is computed by solving at each sampling time an optimization problem which uses the current system state as the initial condition. In general the control move ut at time t, for time invariant systems, is a nonlinear static function of the system state x_t, i.e. u_t = f(x_t). One of the main limitations in using MPC techniques is the presence of fast plant dynamics which require small sampling periods that do not allow to perform the optimization problem online: this motivates the research efforts devoted to develop computationally tractable MPC solutions, or suitable approximations of MPC control laws, by exploiting the properties of f(x). These approaches include explicit piece-wise linear solutions ([2]), neural approximations ([3]) and piece-wise linear approximations ([4]) of f(x). An alternative approach is presented here using Set Membership (SM) methodologies for nonlinear function estimation as firstly introduced in [5]. Under this context, no assumption on the functional form of f(x) is made and only assumptions on its regularity properties, given by bounds on its value set and on its gradient, are considered. The use of such methodology enables to compute an approximation of a given predictive control law which guarantees to fulfill input constraints. Moreover, it is possible to compute an upper bound of the approximation error which holds for every state value in the considered set. In particular, a key feature of this approach is the possibility of tuning the approximation error in order to guarantee satisfaction of state constraints. This way, as the control computation is simply reduced to the evaluation of a static nonlinear function, the computational time is significantly reduced leading to a “Fast” Model Predictive Control implementation (FMPC). Two different SM methodologies are presented for the approximation of the nominal MPC controller. Both methodologies are based on the off-line computation of a certain number N of exact MPC control moves. The first method derives an “optimal” approximating function which minimizes, for a given N, the guaranteed accuracy level. However, its computational time grows with N. The second one gives lower guaranteed accuracy for a given N, but its computational time is lower and it is approximately constant with N. Thus, at the cost of increasing the number of off-line computations, any desired level of guaranteed accuracy can be obtained without increasing the on-line computation effort. Two numerical examples are given to show the effectiveness of the presented results.

REFERENCES
[1] D. Q. Mayne, J. B. Rawlings, C. V. Rao and P.O.M. Scokaert,
“Constrained model predictive control: Stability and optimality”,
Automatica, 36, 789–814, 2000.
[2] A. Bemporad, M. Morari, V. Dua and E.N. Pistikopoulos “The explicit
linear quadratic regulator for constrained systems”, Automatica
38, 3–20, 2002.
[3] T. Parisini and R. Zoppoli, “A receding-horizon regulator for
nonlinear systems and a neural approximation”. Automatica 31(10), 1443–
1451, 1995.
[4] T. A. Johansen, “Approximate explicit receding horizon control of
constrained nonlinear systems”. Automatica 40, 293–300, 2004.
[5] M. Canale and M. Milanese, “FMPC: a fast implementation of model
predictive control”, in 16th IFAC World Congress, Prague, Czech
Republic, July 2005.

Slides

Fri 12 - Fri 12 Oct-07 Doctoral Presentation - Erik Hostens
Auditorium Wolfspoort (Huis Bethlehem), Schapenstraat 34, 3000 Leuven
2:00 pm
"Quantum entanglement distillation in the stabilizer formalism"
Erik Hostens (K.U. Leuven, ESAT-SCD)

Entanglement is a quantum mechanical phenomenon that manifests itself through correlations of local measurements, unexplained in any classical theory (Einstein called it "spooky action at a distance"). Since a few decades there has been a growing research interest in entanglement, concerning both fundamental properties (quantum information theory) and practical applications (such as quantum computing, quantum cryptography and quantum communication).
An important property of entanglement is that it cannot increase by local operations and classical communication only: two systems need to interact physically to become entangled. However, certain applications do not allow for such interactions. But it is possible to concentrate the amount of entanglement in a part of the system by local operations and classical communication. This is called entanglement distillation and the used methods are distillation protocols. They are both of theoretical and of practical importance: on the one hand, they serve to give a better insight into the properties of entanglement and the physical boundaries of entanglement manipulation; on the other hand, pure entanglement cannot be realized in practice because of influences of the environment and noise in the used quantum channels, which makes distillation indispensable.
In this thesis, we develop distillation protocols for bipartite and multipartite entanglement. We apply the stabilizer formalism, a mathematical framework that can be formulated in terms of binary linear algebra. This 'binary picture' allows for a transparent and efficient description of distillation protocols and gives us the opportunity to improve existing results significantly.

Jury:Prof. dr. ir. J. Berlamont, chairmanProf. dr. ir. B. De Moor, promoterDr. ir. J. Dehaene, copromoterProf. dr. M. FannesProf. dr. ir. J. VandewalleProf. dr. ir. F. Verstraete (Universität Wien)Prof. dr. ir. K. Audenaert (University of London)

Mon 8 - Tue 9 Oct-07 Workshop on Optimal Experimental Design
Esat, Room 00.62
9:00 am-6:00 pm

October 8-9, 2007
Katholieke Universiteit Leuven

Aim:
This workshop wants to bring researchers with interests in optimal experimental design (OED) together, that work in various fields of engineering including chemical, biochemical, and mechanical engineering. The workshop is organized in two days.

The first day is a tutorial day introducing the principles of optimal experimental design for nonlinear model identification using a series of examples and software exercises (guest teacher: Stefan Koerkel, Humboldt University Berlin). This day is an excellent opportunity to get acquainted with OED for newcomers, and to learn how to use state-of-the-art software for nonlinear parameter estimation and OED.

The second day will be a scientific day. Several invited speakers will present their work on OED, and there is the possibility to present your own work in form of a talk (limited places) or a poster. In case you are interested, please submit a title and short abstract (plain text, max. 250 words) to Christian Hoffmann.

Confirmed international speakers:

• Stefan Körkel, Institut of Mathematics, Humboldt University Berlin
• Hergen Schultze, BASF AG Ludwigshafen, Germany
• Samuel Bandara, Chemical and Systems Biology, Stanford University, USA.

Wed 3 - Wed 3 Oct-07 K.U. Leuven Seminars on Optimization in Engineering - Daniel Leineweber
00.62
4:00 pm-5:00 pm
"Advanced Modeling Technology - the Key to Optimal Processes"

Daniel Leineweber
Bayer Technology Services GmbH, Leverkusen

Computer based modeling, analysis and optimization techniques have become an essential tool for process improvement in the chemical process industries. In order to successfully tackle real-life problems, it is often necessary to use a combination of several advanced techniques, resulting in a multi-methodic approach. Three practical examples are discussed:

- Data mining analysis of a polymer process (troubleshooting)
- Multi-methodic optimization of a spandex spinning process (sensitivity
reduction)
- Rigorous optimization of a catalytic tube reactor (capacity
maximization)

It is shown that multi-methodic approaches based on data mining, neural networks, rigorous modeling, and various optimization techniques can be successfully applied in practice to better understand and, ultimately, optimize complex processes.

Wed 3 - Wed 3 Oct-07 K.U. Leuven Seminars on Optimization in Engineering - Ian Couchman
01.60
5:00 pm-5:30 pm

“Optimal Control of Mixing”

Ian Couchman (Imperial College London, UK)
E.C. Kerrigan, J.C. Vassilicos

Motivated by the problem of laminar fluid mixing, this presentation will consider the problem of optimal control of advective mixing in a 2D Navier-Stokesfluid flow. The type of flow that is studied here was first introduced in [1]; a charge-carrying flow is induced by a spatially distributed force field, the strength of which can be scaled arbitrarily with a time-varying current. Though the flow is laminar, it has turbulent-like statistics and is well suited to mixing [1].

The control objective that will be considered in this presentation, is how to compute the current profile that will maximize the mixing in the flow for a given amount of input energy. The contribution of this presentation is two-fold:

1. The degree of mixing of a flow will be quantified by considering a number of different cost functions. A new measure, based on the idea that mixing can be quantified by the frequency with which parts of the flow come together, will be introduced and justified.

2. The optimal control problem, which contains a dynamic model for the evolution of the degree of mixing of the fluid, will be defined in detail.

The presentation will conclude with specific research questions regarding the efficiency and reliability of the numerical solution of the optimal control problem.

References

[1] L. Rossi, J.C. Vassilicos, and Y. Hardalupas. Electromagnetically controlled

multi-scale flows. J. Fluid Mech, 2006.

Slides

Wed 26 - Wed 26 Sep-07 Course on "Numerical Optimization" - Moritz Diehl
see schedule
10:35 am-12:35 pm

Lectures on "Numerical Optimization" given by Moritz Diehl taking place first on Sept 26 (until Dec 11.)
every tuesday (room Celestijnenlaan 200 C 01.05, Aud. C) and
wednesday from 10:35-12:35(room Celestijnenlaan 300 - 00.81, Aud. D).

Announcement

Contents of the Course

Part I: Introduction
1.   Fundamental Concepts of Optimization
2.   Types of Optimization Problems

Part II:  Unconstrained Optimization
3.   Optimality Conditions and Convexity
4.   Iterative Descent Methods
5.   Newton Type Optimization Methods
6.   Calculating Derivatives
7.   Estimation and Fitting Problems
8.   Trust Region vs. Line Search Methods

Part III: Constrained Optimization
9.   Karush-Kuhn-Tucker Optimality Conditions
12. Globalisation Strategies
13. Interior Point Methods
14. Optimal Control Problems
15. Summary of the Lecture
16. Duality
17. Special Topics in Convex Optimization

The course is partly based on the book "Numerical Optimization" by
J. Nocedal and S. Wright, Springer Verlag.

Lecture:         Moritz Diehl,
Exercises:       Hans-Joachim Ferreau
Special Topics:  Michel Baes,

Please register by sending an email with the subject NUMOPT
to Ida Tassens

Wed 26 - Wed 26 Sep-07 K.U. Leuven Seminars on Optimization in Engineering - Marnix Volckaert
Esat 91.91
4:00 pm-5:00 pm

“Interplanetary Spacecraft Trajectory Optimization”

Marnix Volckaert (T.U.Delft)

In order to learn more about the origin of our solar system, it is crucial to send spacecraft to other planets and their moons. Furthermore, the possibility of sending humans to another planet, such as Mars, requires the technology to transfer a spacecraft from the Earth to the other planet. Such a complex undertaking can only be achieved if the trajectory followed by the spacecraft is highly optimized, for example in terms of the required propellant mass.

Finding the trajectory that uses the least amount of propellant is difficult. The search space is quasi unlimited, and there are multiple local optima. The motion of the spacecraft is described by complex astrodynamical concepts, that can be used to the advantage, for example during a gravity assist maneuver.

Trajectory optimization techniques can be divided into analytical methods and numerical methods. The analytical methods are generally much faster but can be inaccurate because of their simplifying assumptions. The numerical methods can be more accurate, but are usually slower, because they rely on numerical integration of the spacecraft's state.

In this presentation an overview will be given of different trajectory optimization methods, both analytical and numerical, including their application to real-life and concept missions.

Tue 25 - Tue 25 Sep-07 K.U. Leuven Seminars on Optimization in Engineering - Thomas Davi
Esat 00.62
4:00 pm-5:00 pm

"A worst-case-estimate for the method of conjugate gradients"
Thomas Davi (University of Duesseldorf)

The cg-algorithm is used for quadratic functions whose appropriate symmetric nxn matrix is positive definite. This demand ensures that the minimum of the function is found within n steps (theoretically - because of rounding errors in general n steps are not enough in practice). For a given dimension n and a condition number K>1 we deal with the maximum possible distance between the n-1. iterate of the cg-algorithm and the minimum of any quadratic function whose symmetric postive definite matrix has the condition number K. To measure the distance we construct a function F which contains n-1steps of the cg-algorithm. The input of F is a quadratic function and a starting vector. We analyze the function F and try to find candidates for the maximum of it by a trust-region-method.

Wed 5 - Wed 5 Sep-07 K.U. Leuven Seminars on Optimization in Engineering - Dirk Abel
Esat, Room 00.62
4:00 pm-5:00 pm

“Real-time optimization in automotive control applications”

Ralf Beck, Dirk Abel (IRT, Aachen)

The need for more fuel efficient vehicles is a key driver for innovations in automotive engineering. This goal gives rise to sophisticated combustion engine concepts like Controlled Auto Ignition (CAI) and driveline architectures like Hybrid Electric Vehicles (HEV) which in turn, due to their inherent complexity, require advanced control strategies. This favors the introduction of optimization-based control schemes like Model Predictive Control to these fast applications, which is addressed by this talk.

Besides the control of driveline components, the overall drivetrain energy management gains importance in order to maximize the vehicle’s efficiency and lower th fuel consumption. This naturally leads to an optimization problem. The talk will exemplify this topic by the energy management of hybrid electric vehicles.

Slides

Tue 4 - Tue 4 Sep-07 K.U. Leuven Seminars on Optimization in Engineering - Pierre-Brice Wieber
ESAT, Room 91.91
11:00 am-12:00 pm

"Optimization in the Modeling and Control of Walking Robots "
Pierre-Brice Wieber (INRIA, Grenoble)

The modeling of walking articulated systems introduces two  fundamental questions that deal with the unilateral contact between  the feet of the system and the ground. The first question is about  the global behavior of the system that results from this unilateral  contact, its general stability. The second question focuses more  specifically on the changes in the support phases and their general  analysis. These two topics raise many serious theoretical  mathematical modeling problems, and they deeply impact the control  laws that we would like to implement on these systems to make them  realize stable and robust walking movements. Far from the very  impressive demos of humanoid robots that have been proposed by  japanese industrials during the last years, these are in fact serious  problems to which no satisfying answers exist today. We will propose  in this talk a quick tour of all these questions and some of the  possible answers, with a special emphasis on where, why and how the  tools from the theory of Optimization are a major help.

Slides

Thu 2 - Thu 2 Aug-07 K.U. Leuven Seminars on Optimization in Engineering - Ilario Gerlero, Filiberto Fele
Esat 00.62
4:00 pm-5:00 pm
"KiteGen Project: Optimisation of a working cycle for the Carousel wind energy generator"
Ilario Gerlero and Filiberto Fele (Politecnico di Torino)

Abstract:
The Kite Wind Generators represent a class of wind generators, based
on the control of tethered airfoils (kites) aimed to be able to
overcome the main limitations of the present aeolian technology based
on wind mills.

Two configurations are considered here for wind energy generation,
indicated as push-pull (Yo-Yo) and Carousel, respectively.

The talk presents the Carousel KWG model, highlighting the main
results achieved by the kite’s research group of Prof. Mario Milanese
from “Politecnico di Torino”. The presentation will carry on with the
aim of our work, which consist in the numerical optimisation of  the
working dynamics for the Carousel KWG, in order to maximize the
produced energy. The solution that will be achieved for the Carousel
KWG dynamics will give us the possibility to compare it with the yo-yo
configuration, already optimized. Finally we will show the results
achieved in first two weeks of our study, discussing about the problems
founded until now.

The numerical optimisation is done by using the package MUSCOD-II,
based on Bock’s direct multiple shooting method.

Mon 30 - Mon 30 Jul-07 SISTA Seminar - Sri Priya Ponnapalli
ESAT 00.62
2:00 pm
"Higher-Order Generalized Singular Value Decomposition for
Comparative  Analysis of Large-Scale Datasets"
Sri Priya Ponnapalli (Univ Texas)

Comparative analysis of large-scale data sets promises to enhance our
fundamental understanding of the data by distinguishing the similar
from the dissimilar among these data. Recently we showed that when data
sets are tabulated as matrices, the generalized SVD (GSVD) provides a
comparative mathematical framework for two large-scale data sets. We
now define a higher-order GSVD (HO GSVD) of more than two matrices
having the same number of columns, and show that this HO GSVD provides
a comparative mathematical framework for more than two large-scale
data sets. This HO GSVD extends to higher-order most of the
mathematical properties of GSVD. We illustrate the  framework offered
by HO GSVD with a comparison of three genome-scale mRNA expression
data sets from three different organisms, human, the yeast
Saccharomyces cerevesiae, and the yeast Schizosaccharomyces pombe,
during their cell-cycle.

Thu 26 - Thu 26 Jul-07 K.U. Leuven Seminars on Optimization in Engineering - Jorge Nino Castaneda
Esat 00.62
4:00 pm-5:00 pm

"Model Identification of an UAV in straight steady flight condition"
Jorge Niño
Universiteit Gent, België

The thesis work is focused on the model identification of an UAV (unmanned aerial vehicle) in straight steady flight condition, based on input-output data collected from flight tests using both frequency and time domain techniques. The vehicle used for this work is an in-house 40 cm wingspan airplane. Because of the complex coupled, multivariable and nonlinear dynamic of the aircraft, linear SISO structures for both the lateral and the longitudinal model around a reference state were derived. The aim of the identification is to provide models that can be used in future UAV applications that involves the implementation of control techniques.

Basically the presentation is about general models for UAVs, linear models for flight conditions, hardware and software designed for model identification experiments and results regarding a specific vehicle: The 40 cm wingspan UAV designed and built in UGent. Some references about possible control strategies are given.

Tue 24 - Tue 24 Jul-07 K.U. Leuven Seminars on Optimization in Engineering - Toshiyuki Ohtsuka
Esat 00.62
4:00 pm-5:00 pm

"Continuation/Krylov Method and Multiple Shooting for a Real-Time Algorithm of  Nonlinear Model Predictive Control"
Prof. Toshiyuki Ohtsuka (Osaka University)
http://www-sc.sys.es.osaka-u.ac.jp/~ohtsuka/

Abstract: This talk gives an overview of continuation-based real-time
algorithm for nonlinear model predictive control. The optimal solution is
updated by solving a linear equation only once at each sampling time, and
the linear equation can be solved efficiently with a Krylov subspace method,
GMRES. In addition to the basic idea of the algorithm, implementation
examples for mechanical systems are also presented with emphasis on
practical handling of inequality constraints. Inequalities are treated by
the introduction of penalized slack variables in a way that bears some
similarity to interior point methods with a fixed barrier parameter. For
further improvement of accuracy and computational time of the algorithm,
multiple shooting and condensing are also introduced. This is a joint work
with Moritz Diehl  and a former student, Yuichi Shimizu.

Fri 20 - Fri 20 Jul-07 ICIAM Minisymposium "Interior Point Methods for MPC" - Zurich
Zurich
11:15 am-4:15 pm
Minisymposium organized by Moritz Diehl (K.U. Leuven, Belgium) and
Steve Wright (University of Wisconsin, USA) as part of the International Congress on Industrial and Applied Mathematics (ICIAM07). Conference website. Find the presentations of the minisymposium here.

Fri 13 - Fri 13 Jul-07 K.U. Leuven Seminars on Optimization in Engineering - Bharath Rangarajan
Esat 00.62
11:00 am-12:30 pm
"Decomposition methods for large Linear Matrix Inequalities"

Prof. Dr. Bharath Rangarajan
Department of Mechanical Engineering,
University of Minnesota (USA)

The stability conditions for spatially-distributed control systems give rise
to large semidefinite programs with specific structure. The standard
gradient techniques are usually based on the Dantzig-Wolfe decomposition. We
explore related smooth counterparts that can be solved using interior-point
techniques which promise faster asymptotic convergence as well as an
improvement in global convergence. There are special properties that are
present in use of gradient techniques that help distribute the computation.
This feature is absent from interior-point methods. The ideas of
decomposition using ipms have also be studied in the context of stochastic
programs.

Mon 9 - Mon 9 Jul-07 Simon Stevin Lecture on Optimization in Engineering - Stephen Wright
Auditorium of the Arenberg Castle
5:00 pm-6:00 pm

Third Simon Stevin Lecture on Optimization in Engineering  (flyer)

" Applying Optimization in Science and Engineering "
Stephen Wright (University of Wisconsin-Madison, USA)

Abstract:                       Slides,   pictures

Much of the vitality for modern optimization research is drawn from its application to research problems in other areas. Collaborations of this type are driving research on new algorithms and are causing previously known algorithmic tools to be reevaluated and recombined. This talk addresses both technical and "human" aspects of collaborations between optimization specialists and other scientists and engineers.
I discuss several interesting applications from my own experience, including signal and image processing, logistic regression, model predictive control, and cancer treatment planning. Issues that arise frequently in these and other modern large-scale applications will be highlighted, and algorithmic approaches will be discussed. I conclude with some observations about the technical background and state of mind required to establish successful collaborative relationships with applications scientists.

Bibliographical Information:
Stephen J. Wright is a Professor in the Computer Science Department at the University of Wisconsin-Madison. Among other, he is Chair-Elect of the Mathematical Programming Society and  Editor in  Chief of the Mathematical Programming Journal (Series B). His research focus is on numerical optimization - theory, algorithms, and implementations - and in applications of all types.

Professor Stephen Wright obtained his Ph.D. in  1984 from the University of Queensland, Australia, and held positions at the University of Arizona and North Carolina State before becoming a computer scientist at the famous Argonne National Laboratory in 1990. In 2001 he moved to the University of Wisconsin-Madison. Professor Wright is the author of a variety of well cited journal articles and two excellent and popular optimization textbooks: Primal-Dual Interior-Point Methods (1997) and Numerical Optimization (with Jorge Nocedal, Second edition August 2006).

The "Simon Stevin Lecture Series on Optimization in Engineering" is set up in order to promote optimization in engineering. For this aim, every quarter of the year an outstanding international scholar is invited to report on latest progress in the development of optimization algorithms and their applications in engineering.
Simon Stevin (1548-1620) was a Flemish mathematician and engineer. Among other, he helped to advance the use of decimal fractions, was the first to explain the tides by the attraction of the moon, and discovered the hydrostatic paradox. He made numerous inventions, among them a wind propelled carriage with sails, the "land yacht", which once impressed Prince Maurice of Orange as it moved faster than horses, in around 1600 on the beach between Scheveningen and Petten. Simon Stevin was fond of promoting the use of science in daily life and in craftmanship, and translated various mathematical terms into dutch. Among other, he introduced the dutch word for mathematics, "wiskunde".

Directly after this Summer's Simon Stevin Lecture, a little reception will be given at 18:00 in the salons of Arenberg Castle, to which all attendants of the lecture are most warmly welcome!

***** REGISTRATION ENCOURAGED *****
Please send an e-mail with the subject "STEVIN" to Ida.Tassens@esat.kuleuven.be if you intend to participate in the event. No obligation, just to help us getting an idea how many people plan to come.

This Stevin lecture is co-sponsored by ICCoS (Identification and Control of Complex Systems), a Scientific Research Network of the Research Foundation - Flanders (FWO-Vlaanderen).

Mon 9 - Fri 13 Jul-07 Course on Convex Optimization - Michel Baes, Bharath Rangarajan
Esat, Room 00.62
9:00 am-12:30 pm

"Course on Convex Optimization"
Michel Baes (K.U.Leuven) and Bharath Rangarajan (University of Minnesota, USA)

The difference between an efficiently tractable optimization problem and an extremely difficult one is sometimes not easy to see for a non-expert. However, this issue is of crucial importance for practitioners. If their problem is poorly modeled, the chance that the optimization software they use gives a reliable solution is close to zero.

Convex optimization problems consitute an important tractable class of instances. The purpose of this course is to give practitioners some tools to model some real-life problems as convex optimization problems, and to give them an insight on the algorithms used to solve them efficiently. The first part of the course will provide a review of basic convexity theory and of conic duality theory. Next, we will focus on some important classes of convex problems, including Second-Order Cone Programming, Semidefinite Programming, and Geometric Programming. Several application examples will be discussed and some modeling techniques will be described, such as the S-Lemma and Schur's Lemma.

All these problems can be solved very efficiently both in theory and in practice. We will review two important classes of algorithms, the Black-Box Methods and the Interior-Point Methods, and explain what makes them so efficient.

Finally, we will cover some newer topics, related to decomposition techniques of extremely large problems and using interior-point methods to detect infeasibility.

References
[1] Yurii Nesterov, Introductory lectures on Convex Optimization: A Basic course, Springer.
[2] Stephen Boyd and Lieven Vandenberghe, Convex Optimization, Cambridge University Press.
[3] Imre Polik and Tamas Terlaky, Survey of S-Lemma, SIAM Review.
[4] Aaron Ben-Tal and Arkadi Nemirovski, Lectures on Modern Convex Optimization: Analysis, Algorithms, and Engineering Applications, SIAM.

Schedule
Mon. (slides)
9h00-10h30:

• Convexity: definitions, examples.
• Optimization: examples and models for convex optimization.
11h00-12h30:
• Optimality conditions
• Easy and intractable problems.
Tue.
9h00-10h30: (slides)
• Duality theory
• Conjugate functions
11h00-12h30: (slides)
• Introduction to optimization methods.
• Black-box methods
Wed.
9h00-10h30: (slides)
• Interior-point methods for conic programming
11h00-12h30: (slides)
• Interior-point methods (cont.)
• Dealing with infeasibility
Thu.
9h00-10h30: (slides)
• Second-order, semidefinite representability
• S-Lemma and applications

11h00-12h30 (slides)

• Convex modeling: applications and demonstrations.
afternoon: (exercise)
• Computer session: using Sedumi
Fri. 13 (Special topics)
9h00-10h30: (slides)

• Smoothing techniques
11h00-12h30: (slides)

• Decomposition methods for very large problems

***** REGISTRATION *****
If you want to attend the course, please send an e-mail to Ida.Tassens@esat.kuleuven.be before July 3. Please indicate in your email wether you want to get an evaluation or not.

This course on convex optimization is funded by OPTEC and by ICCoS (Identification and Control of Complex Systems), a Scientific Research Network of the Research Foundation - Flanders (FWO-Vlaanderen).

Mon 2 - Wed 4 Jul-07 SynCoNet2007 - International Symposium on Synchronization in Complex Networks
Leuven Belgium

Wed 27 - Wed 27 Jun-07 Minicourse on Dynamic Optimization with MUSCOD-II
ESAT Room 2.54
2:00 pm-7:00 pm

"Minicourse on Dynamic Optimization with MUSCOD-II"

Teachers:
Hans Joachim Ferreau, Julian Bonilla,
Moritz Diehl, Joris Vanbiervliet

Abstract: Aim of this 5 hour course is to provide working knowledge of the software package MUSCOD-II for numerical solution of nonlinear optimal control problems. We start by formulating a simulation example, and then set up tutorial optimal control problems. Towards the end of the course, each participant is encouraged to start formulating his/her own example. Basic knowledge of LINUX, a suitable editor, and the C-programming language is required.

#### Supporting material:

Handouts

Original paper by Bock, 1984

Overview paper by Binder et al, 2001

Wed 27 - Wed 27 Jun-07 K.U. Leuven Seminars on Optimization in Engineering - Knut Graichen
Esat 02.58
10:00 am-11:00 am

"Constrained feedforward control design and some first extensions to optimal control"
Knut Graichen
Postdoc at Centre Automatique et Systèmes, Ecole des Mines de Paris(France), on indirect optimal control (joint work with Nicolas Petit)

Feedforward controls are used in many practical control applications to enhance the tracking performance while maintaining the classical feedback loop. A typical control problem thereby is the transition between stationary setpoints, e.g. position changes of mechatronic systems or load changes in process control. In this context, the first part of the talk presents a new approach of inversion-based feedforward control design, which treats the setpoint change scenario as a two-point boundary value problem (BVP) in the coordinates of the input-output normal form. Constraints on the output trajectory and its time derivatives are directly incorporated within the formulation of the BVP by means of saturation functions. The approach is illustrated for swing-up and side-stepping maneuvers of a double and triple pendulum with both numerical and experimental results.
The second part of the talk presents first results on how the saturation function concept for handling the output constraints in the feedforward design can be extended to optimal control problems with state and input constraints. Several examples are used to illustrate the ideas.

Slides

Thu 21 - Thu 21 Jun-07 K.U. Leuven Seminars on Optimization in Engineering - Fred Hamprecht
ESAT, Room 00.62
4:00 pm-5:00 pm

"Exploiting spatial context in the quantification of vector valued magnetic resonance data: a high-dimensional optimization problem "
B. Michael Kelm, Björn H. Menze, Fred A. Hamprecht (University of Heidelberg)

Conventional imagery is increasingly complemented by spatially resolved vectorial or tensorial measurements that can help elucidate phenomena that would otherwise elude an unequivocal interpretation. While such data (e.g. spectroscopic images, image sequences, diffusion tensor images) potentially offer valuable additional information, they also pose difficult problems in signal processing, pattern recognition and optimization.

More specifically, and restricting this discussion to medical in vivo magnetic resonance (MR) studies, conventional MR images offer morphological information only. In recent years, new flavors of MR imagery have emerged, such as functional MR imaging for the study of brain activity, diffusion tensor imaging for the elucidation of tissue microstructure and brain connectivity, diffusion contrast enhanced (DCE) imaging for the derivation of pharmacokinetic parameters, or MR spectroscopic images (MRSI) for tumor detection and localization.

Our talk will focus on noisy DCE and MRSI images, and discuss strategies to incorporate spatial context in their analysis. In both cases, a parametric model is fit to the raw data. A spatial smoothness assumption is embodied in the potentials of a Markov Random Field, and the maximum a posteriori estimate is approximated by means of a block-iterated conditional modes algorithm. This heuristic for solving a high-dimensional optimization problem is shown to converge faster than conventional iterated conditional modes (ICM) and to produce parameter estimates with significantly smaller variance, leading to an overall reduction in mean squared error.

Tue 19 - Tue 19 Jun-07 Workshop on "PDE optimization and Inverse Problems" - W. Bangerth, S. Vandewalle, B. Vandereycken
ESAT, Room 00.62
2:00 pm-6:30 pm
Workshop website

Tue 19 - Tue 19 Jun-07 K.U. Leuven Seminars on Optimization in Engineering - Wolfgang Bangerth
Esat - Room 00.62
5:30 pm-6:30 pm

"Numerics for inverse problems in biomedical imaging"
Wolfgang Bangerth (Texas A&M University, USA)

In many of the modern biomedical imaging modalities, the measurable signal
can be described as the solution of a partial differential equation that
depends nonlinearly on the tissue properties (the "parameters") one would
like to image. Consequently, there are typically no explicit solution
formulas for these so-called "inverse problems" that can recover the
parameters from the measurements, and the only way to generate body images
from measurements is through numerical approximation.

The resulting parameter estimation schemes have the underlying partial
differential equations as side-constraints, and the solution of these
optimization problems often requires solving the partial differential
equation thousands or hundred of thousands of times. The development of
efficient schemes is therefore of great interest for the practical use of
such imaging modalities in clinical settings.

In this talk, the formulation and efficient solution strategies for such
inverse problems will be discussed, and we will demonstrate its efficacy
using examples from our work on Optical Tomography, a novel way of imaging
tumors in humans and animals. The talk will conclude with an outlook on
even more complex problems that attempt to automatically optimize
experimental setups to obtain better images.

This seminar is part of a workshop on "PDE Optimization and Inverse Problems", more information.

Thu 14 - Thu 14 Jun-07 K.U. Leuven Seminars on Optimization in Engineering - Francois Glineur
Esat 00.62
3:30 pm-5:00 pm

"Solving a large class of structured convex optimization
problems with a unified conic formulation"

*** The talk will start at 16:00. Before (starting at 15:30) there is a coffee break and the opportunity to meet and have discussions. ***

(joint work with Robert Chares)

The first part of this talk will briefly survey the field of
convex optimization (a subset of nonlinear optimization), whose
main paradigm is the following: in exchange for voluntarily
restricting your optimization model to specific (convex)
objective functions and (convex) feasible sets, you get a
better understanding of your problem (using the notion of
duality) and the ability to use efficient algorithms, both from
the theoretical (algorithmic complexity) and practical (ability
to solve large-scale problems efficiently) points of view. In
addition, we will focus on the necessity to consider structured
problems, and in particular on the conic formulation of convex
problems.

The second part of this talk will describe a common framework
recently introduced to unify several classes of structured
convex optimization problems, including linear programming,
second-order cone programming, quadratically constrained convex
quadratic programming, l_p programming, minimization of sums of
Euclidean or p-norms, geometric programming, entropy
programming, etc.

Each of these problems can be modelled as a conic optimization
problem, where all the cones used in the formulation belong to
a single family of three-dimensional self-dual convex cones,
defined as { (x,y,z) | x^a y^(1-a) >= |z|, x>=0, y>=0 } where a
is a real parameter between 0 and 1.

This unified formulation features several advantages, including
the possibility to effortlessly derive the dual of any of these
problems and the ability to apply the same algorithmic
framework to any of those classes of structured convex
problems, more specifically interior-point methods based on the
theory of self-concordant barriers.

The presentation of a solver currently under development will
conclude the talk.

Slides

Wed 13 - Wed 13 Jun-07 K.U. Leuven Seminars on Optimization in Engineering - Alain Vande Wouwer
00.62
3:00 pm-4:00 pm

"Modelling and identification from batch experiments of a SMB process"
Alain Vande Wouwer (FPMs Mons)

The Simulated Moving Bed (SMB) technology is a continuous chromatographic process which is important in various fields, from sugar to enantiomer separation. In this paper, a systematic identification procedure for determining parameters of SMB models from batch experiments is validated with experimental SMB data. Parameters are first estimated from elution peaks. Then a cross-validation with SMB experiments is performed so as to assess whether the parameters identified from batch experiments may be used in a SMB model. This part of the work requires a careful modelling of the dead volumes within the SMB process.

Mon 11 - Mon 11 Jun-07 Doctoral Presentation - Samuel Xavier de Souza
room Nicolas de le Ville at the Campus Arenberg Library, Willem de Croylaan 6, 3001 Heverlee
2:00 pm

## Optimisation and Robustness of Cellular Neural Networks

### Samuel Xavier de Souza (K.U. Leuven, ESAT-SCD)

Abstract:
In this thesis we present new methodologies for improving the robustness of analog VLSI visual processors which are based on Cellular Neural Networks (CNN). Such a system can process information at very high speeds, only comparable to todayâ€™s supercomputers. The regular lattice architecture of CNNs allows massive parallelism, which makes it very suitable for performance-demanding applications in image processing. Its reduced size and power consumption make it easy to embed in portable appliances. The only disadvantage of todayâ€™s programmable CNNs relates to the analog VLSI technology, which despite remarkable recent advances, cannot guarantee a sufficiently high accuracy and reliability that is needed in many applications. In this thesis we describe methodologies for customised tuning of CNN chips and learning of new complex operations. Based on well established methods such as design centring and trajectory learning, the techniques described here prove to be very useful in reducing the effects of parameter deviation and post-manufacturing interference in the operation of CNN-based processors. We show that on-chip learning is not only viable but also better than simulation-based learning for being much faster. We also present a new global optimisation method that is suitable for CNN optimisation. The new method of Coupled Simulated Annealing makes use of coupling in order to allow multiple Simulated Annealing (SA) processes to cooperate toward finding the global optimum of multi-modal and multidimensional optimisation problems. A number of proof-of-concept applications is presented in order to show the effectiveness of our methodologies. These applications serve to demonstrate the potential for future VLSI CNN systems toward ultra-fast visual applications such as quality control in agricultural, semiconductors, textile and other industries, surveillance and traffic analysis, biochemical process inspection, intelligent systems in the automotive industry, visual computer/gaming interaction and others.

Promotors:
Prof. Johan Suykens
Prof. Joos Vandewalle

Wed 30 - Fri 1 Jun-07 CEA-EDF-INRIA School - INRIA Rocquencourt (France)
INRIA Rocquencourt
9:00 am
Conference website

Wed 23 - Wed 23 May-07 K.U. Leuven Seminars on Optimization in Engineering - Sven Erb (ESA)
ESAT, Room 00.62
2:00 pm-3:00 pm

" Spacecraft Trajectory Optimization in a Project Environment "
Sven Erb (ESA - TEC/ECM , Netherlands)

It is one of the core competencies of the Dynamics and Mathematical
Analysis Unit TEC/ECM at the European Space Agency ESA to conduct
trajectory planning and related analysis services for ESA projects. The
activities range from advanced launch vehicle ascent and re-entry
trajectories and sizing of the vehicle, to spacecraft planetary transfers,
and interplanetary travel. It is common practice to employ optimization
techniques for such tasks. The preferred approach at TEC/ECM for solving
such problems is based on non-linear programming. The standard software is
GESOP/ASTOS, a graphical environment for simulation and optimization that
features NLP solvers, such as SNOPT and SOCS.

The primal problems with continuous controls and optimizable parameters is
transcribed into non-linear programming problems and subsequently solved
by means of NLP solvers. A typical example is the re-entry of a winged
vehicle into the earth atmosphere, where the angle of attack and the
sideslip angle are optimizable.

From the perspective of a practitioner, who uses the optimization software on heterogeneous tasks, robustness, flexibility, user friendliness and
precision are driving criteria for the usefulness of the algorithm. It is
an asset, if the user can stay close to the real world during the
optimization, with established physical properties, and does not have to
deal with costates, arising from indirect solution methods.

The range of optimization problems and sizes is illustrated through a
number of current spacecraft projects. Typical missions are, for instance,
launch vehicles, where the ascent trajectory has to satisfy a multitude of
constraints with very different character; or low-thrust transfer
trajectories, where the problem description can easily go beyond 100.000
parameters.

Tue 22 - Tue 22 May-07 Doctoral Presentation - Frizo Janssens
Auditorium of the Arenberg Castle
1:30 pm
Clustering of scientific fields by integrating text mining and bibliometrics

Frizo Janssens (K.U. Leuven, ESAT-SCD)

Abstract

Increasing dissemination of scientific and technological publications via the Internet, and their availability in large-scale bibliographic databases, has led to tremendous opportunities to improve classification and bibliometric cartography of science and technology. This metascience benefits from the continuous rise of computing power and the development of new algorithms. Paramount challenges still remain, however.

This dissertation verifies the hypothesis that accuracy of clustering and classification of scientific fields is enhanced by incorporation of algorithms and techniques from text mining and bibliometrics. Both textual and bibliometric approaches have advantages and intricacies, and both provide different views on the same interlinked corpus of scientific publications or patents. In addition to textual information in such documents, citations between them also constitute huge networks that yield additional information. We incorporate both points of view and show how to improve on existing text-based and bibliometric methods for the mapping of science.

The dissertation is organized into three parts:

Firstly, we discuss the use of text mining techniques for information retrieval and for mapping of knowledge embedded in text. We introduce and demonstrate our text mining framework and the use of agglomerative hierarchical clustering. We also investigate the relationship between the number of Latent Semantic Indexing factors, the number of clusters, and clustering performance. Furthermore, we describe a combined semi-automatic strategy to determine the optimal number of clusters in a document set.

Secondly, we focus on analysis of large networks that emerge from many individual acts of authors citing other scientific works, or collaborating in the same research endeavor. These networks of science and technology can be analyzed with techniques from bibliometrics and graph theory in order to rank important and relevant entities, for clustering or partitioning, and for extraction of communities.

Thirdly, we substantiate the complementarity of text mining and bibliometric methods and we propose schemes for the sound integration of both worlds. The performance of unsupervised clustering and classification significantly improves by deeply merging textual content of scientific publications with the structure of citation graphs. Best results are obtained by a clustering method based on statistical meta-analysis, which significantly outperforms text-based and citation-based solutions.

Our hybrid strategies for information retrieval and clustering are corroborated by two case studies. The goal of the first is to unravel and visualize the concept structure of the field of library and information science, and to assess the added value of the hybrid approach. The second study is focused on bibliometric properties, cognitive structure and dynamics of the bioinformatics field. We develop a methodology for dynamic hybrid clustering of evolving bibliographic data sets by matching and tracking clusters through time.

To conclude, for the complementary text and graph worlds we devise a hybrid clustering approach that jointly considers both paradigms, and we demonstrate that with an integrated stance we obtain a better interpretation of the structure and evolution of scientific fields.

Promotors
• Prof. dr. ir. B. De Moor
• Prof. dr. ir. K. Debackere

Tue 15 - Tue 15 May-07 SISTA Seminar - Olivier Roy
ESAT 01.60
3:00 pm

"Collaborating Hearing Aids: An Information-Theoretic Perspective"
Olivier Roy (EPFL)

We investigate the beamforming gain provided by hearing aids that are allowed to collaborate using a rate-constrained wireless link and thus benefit from the signals recorded at both ears of the user.
As a means to benchmark the performance of any practical scheme, we look at the information-theoretic optimal tradeoff between the communication bit-rate (measured in kb/s) and the gain (measured in dB) achieved by this collaboration.

In this talk, I will first state the problem of collaborating hearing aids. Then, some basic concepts of information theory will be presented as a means to understand the framework in which we are working.
Finally, various tradeoffs between communication bandwidth and beamforming gain will be discussed along with their corresponding rate allocation strategy. Some insights about practical implementation will also be given.

This is joint work with Martin Vetterli.

Tue 15 - Tue 15 May-07 K.U. Leuven Seminars on Optimization in Engineering - Georges Gielen
ESAT Auditorium A
4:00 pm-5:00 pm

"Structural synthesis of analog electronic circuits via genetic algorithms "
Georges Gielen (KULeuven - ESAT - Micas)

The design of analog electronic circuits is a very tedious process that requires years of accumulated experience. The current pace of innovations in the field however demands faster design times, hence the need for design automation. After having solved the problem of optimally sizing basic blocks, the next challenge is to also automate the structural synthesis of the circuits. An approach based on genetic algorithms will be presented that enables to automatically explore a large space of circuit topologies.

Thu 3 - Thu 3 May-07 SISTA Seminar - Ivan Goethals
ESAT 00.62
4:00 pm

"Engineering in financial risk management"
Ivan Goethals (Group Credit Portfolio Management - Fortis Group - Brussels)

In order to avoid bankruptcy due to unforeseen adverse movements in
the market, banks have been heavily investing in financial risk
management over the last few years.  Banks can limit the risk of
suffering large losses by steering the so-called intake; a process
that is mostly involved with limiting the number of clients that might
for instance face difficulties with the repayment of their loans.
Nevertheless, even the best selection criteria would not prevent a
massive flooding in, say, the Netherlands from leading to serious
losses in the roughly 60 BN dutch mortgage portfolio of a bank like
Fortis.

To manage these so called 'systemic' risks affecting many clients at
once, banks make increasing use of recent financial innovations which
enable them to physically or virtually sell part of a portfolio that a
bank deems too large or dangerous to other global financial players,
and replace it with a different kind of risk that the bank is not yet
exposed to.  The final aim is to arrive at a portfolio that is as
diversified as possible so that problems in certain economic areas
only affect a limited part of the bank's portfolio.

The process of determining what should be bought and sold and under
what conditions is largely steered by mathematical models.  From
analytical expressions deriving a fair price for the more common
financial instruments to a brute-force simulation of potential future
states of the world economy and their impact on the portfolio,
mathematicians, physicists and engineers play an increasingly
important role in the banking industry.  In this seminar we will
highlight that role and introduce some of the more commonly used
methodologies in a bank like Fortis.  As a case study we will zoom in
on the recent house price boom seen throughout the western world and
elaborate on the role risk transfer mechanisms played in this.

Wed 2 - Wed 2 May-07 K.U. Leuven Seminars on Optimization in Engineering - Carlos Dorea
ESAT 01.57
11:00 am-12:00 pm

Design of set-invariant observers for linear discrete-time systems

Carlos E. T. Dorea

The concept of set-invariance has been intensively used the last years to solve control problems with constraints. In particular, if the constraints are given by a set of linear inequalities, it is possible to construct a controlled invariant set contained in the polyhedron defined by the constraints, such that a suitable sequence of control inputs can be computed to enforce the constraints along the state trajectory.

Based on this concept, we recently proposed a solution for the dual problem, i.e. the design of full-order state observers with limitation of the estimation error. Conditions were established under which a given polyhedral set defined on the estimation error space is invariant, in the sense that the error trajectory can be kept in this set by means of a suitable output injection. Then, we addressed the problem of computing an invariant polyhedron which bounds as much as possible the trajectory of the estimation error, given a polyhedral set of possible initial states.

In this talk, we present the basic technique and discuss some important issues, such as convergence and computational effort of the proposed algorithms, minimality of the computed invariant sets and the computation of the output injection. We conclude by discussing the potential application of the results to the solution of control problems with constraints via output feedback.

Slides

Thu 19 - Fri 20 Apr-07 NMPC-SOFAP 2007: Workshop on Nonlinear Model Based Control - Software and Applications
Loughborough, UK

The aim of the two days workshop is to bring together scientists and practitioners working on nonlinear model based predictive control (NMPC) in real world applications, and on software solutions that broaden the applicability of this powerful control approach.

Workshop website

Wed 18 - Wed 18 Apr-07 Simon Stevin Lecture on Optimization in Engineering - Stephen Boyd
Auditorium Arenberg Kasteel
11:00 am-12:00 pm

Second Simon Stevin Lecture on Optimization in Engineering

"Convex Optimization"
Stephen Boyd (Stanford University, USA)

Abstract:                       Slides

Joint work with Lieven Vandenberghe and Michael Grant

In this talk I will give an overview of general convex optimization, which can be thought of as an extension of linear programming, and some recently developed subfamilies such as second-order cone, semidefinite, and geometric programming.  Like linear programming, we have a fairly complete duality theory, and very effective numerical methods for these problem classes; in addition, recently developed software tools considerably reduce the effort of specifying and solving convex optimization problems. There is a steadily expanding list of new applications of convex optimization, in areas such as circuit design, signal processing, statistics, machine learning, communications, control, finance, and other fields.  Convex optimization is also emerging as an important tool for hard, non-convex problems, where it can be used to generate lower bounds on the optimal value, and as a heuristic method for generating suboptimal points.

Bibliographical Information:
Stephen P. Boyd is the Samsung Professor of Engineering, and Professor of Electrical Engineering in the Information Systems Laboratory at Stanford University. His current research focus is on convex optimization applications in control, signal processing, and circuit design.

The "Simon Stevin Lecture Series on Optimization in Engineering" is set up in order to promote optimization in engineering. For this aim, every quarter of the year an outstanding international scholar is invited to report on latest progress in the development of optimization algorithms and their applications in engineering.
Simon Stevin (1548-1620) was a Flemish mathematician and engineer. Among other, he helped to advance the use of decimal fractions, was the first to explain the tides by the attraction of the moon, and discovered the hydrostatic paradox. He made numerous inventions, among them a wind propelled carriage with sails, the "land yacht", which once impressed Prince Maurice of Orange as it moved faster than horses, in around 1600 on the beach between Scheveningen and Petten. Simon Stevin was fond of promoting the use of science in daily life and in craftmanship, and translated various mathematical terms into dutch. Among other, he introduced the dutch word for mathematics, "wiskunde".

Directly after this spring's Simon Stevin Lecture, a little reception will be given at 12:00 noon in the salons of Arenberg Castle, to which all attendants of the lecture are most warmly welcome!

***** REGISTRATION ENCOURAGED *****
Please send an e-mail with the subject "STEVIN" to Ida.Tassens@esat.kuleuven.be if you intend to participate in the event. No obligation, just to help us getting an idea how many people plan to come.

This Stevin lecture is co-sponsored by ICCoS (Identification and Control of Complex Systems), a Scientific Research Network of the Research Foundation - Flanders (FWO-Vlaanderen).

Thu 5 - Thu 5 Apr-07 SISTA Seminar - Toni Barjas-Blanco, Jeroen Boets
ESAT 00.62
4:00 pm

"Flood prevention in the Demer bassin using Model Predictive Control"
Toni Barjas-Blanco (K.U. Leuven, ESAT-SCD)

In the last decennium the Demer bassin has been confronted with several heavy flooding events. These flooding events caused a lot of troubles in the areas located in the bassin. In order to prevent flooding two big reservoirs are available which are controlled by a simple three-position controller. However, simulations have shown that control actions different than the ones obtained from the three-position controller could have significantly or even completely reduced flooding in the past. Therefore the decision was made to replace the current controller. In this presentation the three-position controller will be replaced by a model predictive controller (MPC). Some problems typically for this application will be discussed as well as their solution. The MPC will be validated on historical test data and its performance will be compared with that of the three-position controller.

"Subspace angles in linear stochastic processes"Jeroen Boets (K.U. Leuven, ESAT-SCD)In this presentation we consider the use of angles in the context oflinear stochastic processes (ARMA processes). A generalization of theangle between two vectors leads to the concept of principal anglesbetween two subspaces, the statistical counterpart of which is canonical correlation analysis (CCA). CCA is a statistical tool for measuring the linear relationship between two random vectors and generalizes the usual correlation between two scalar random variables. For two jointly Gaussian random variables there also exists a relation between the canonical correlations and the mutual information of the two variables.These notions will be applied in the context of linear stochasticprocesses in order to quantify the amount of dynamics in a process, andto define so-called angles between two processes. For scalar processesthese definitions lead to a norm and a distance for ARMA processes. Inthe talk we also demonstrate the practical use of this distance.

Mon 19 - Fri 23 Mar-07 ATHENS course "Dynamic Process and System Optimization" - Moritz Diehl, Jan Albersmeyer
ESAT

This course is directed to master and PhD students both from OPTEC and from partner universities within the ATHENS programme.

ATHENS Course KUL6 "Dynamic Process and System Optimization"

Teachers: M. Diehl, J. Albersmeyer

Fri 2 - Fri 2 Mar-07 K.U. Leuven Seminars on Optimization in Engineering - Walter Gomez
Plantkunde, Room 01.30, Kasteelpark Arenberg 31, Heverlee
11:00 am-12:00 pm

"A filter algorithm for nonlinear semidefinite programming"
Dr. Walter Bofill (Chile)

In this work a filter method for solving nonlinear semidefinite programming problems (NLSDP) is proposed.  The method extends to this setting (NLSDP) the filter sequential quadratic programming algorithm.  This filter approach was recently introduced for the case of nonlinear programming (NLP).  The main result is a proof for the global convergence of the algorithm.  Moreover we discuss our numerical experiences using some problems from the publicly available benchmark collection COMPleib.

Slides

Thu 1 - Thu 1 Mar-07 K.U. Leuven Seminars on Optimization in Engineering - Yurii Nesterov, Florian Jarre
ESAT 01.57
4:00 pm-6:00 pm

16.00h: "Second-order methods with provable global complexity"
Yurii Nesterov (UCL CORE)

In this talk, we discuss a recent progress in the general second-order
minimization schemes related to the cubic regularization of the Newton's
method. For convex case, we present an accelerated multistep version of the method.
We consider the extensions of the new schemes onto constrained problems.
Preliminary computational results are also discussed.

Slides

17.00h: "An augmented primal-dual method for linear conic minimization"
Florian Jarre (University of Duesseldorf),  Franz Rendl  (University of Klagenfurt)

We present a new iterative method for solving linear minimization problems
over convex cones. The problem is reformulated as an unconstrained problem of
minimizing a differentiable convex function. The method does not use any
homotopy parameter but solves the primal-dual problem in one step using a
nonlinear conjugate gradient type approach. In the case of a semidefinite program
we  propose a regularizing function that makes the generalized Hessian positive
definite when there is a unique and strictly complementary optimal solution.
Numerical examples for some classes of difficult semidefinite programs illustrate
the potential of the new approach.

Slides

Wed 21 - Wed 21 Feb-07 K.U.Leuven Seminars on Optimization in Engineering - Peter Kühl
ESAT 02.58
4:00 pm-5:00 pm

16.00h : "Towards Robust Nonlinear Constrained Model Predictive Control"
Peter Kühl, University of Heidelberg

The talk presents an overview over recent developments in the field of nonlinear model predictive control (NMPC) achieved in the simulation and optimization group at the IWR (Interdisciplinary Center for Scientific Computing).  The talk consists of two parts.

First, the advantages of NMPC over standard control schemes are briefly revisited.  Successful NMPC, however, stands and falls with the underlying model and the availability of the entire current state of the process.
The latter problem is addressed by the moving horizon estimator (MHE), a state and parameter estimation approach very similar to NMPC itself.  It is shown how to efficiently treat the optimization problems associated to NMPC and MHE.
Stability is an important issue for both closed-loop control and estimation schemes.  It becomes more complicated once the numerical solution scheme is explicitly taken into account.  Stability of the combined system-optimizer dynamics has recently been shown for both NMPC and MHE.

A way to address the problem of imperfect models wil be subject to the second part of the talk.  For the case of parametric uncertainty, a robust reformulation of the control problem is suggested.  While the approach yields very nice results in the open-loop case, problems arising from closing the loop remain an open topic for research and shall be discussed at the end of the talk.

Slides

Thu 15 - Thu 15 Feb-07 SISTA Seminar - Sergios Theodoridis, Carlos Alzate
ESAT 00.62
3:00 pm-5:00 pm
15.00h: "Support vector machines: a geometric point of view"
Sergios Theodoridis (University of Athens, Greece)

Support Vector Machines have been established as one of the major classification and regression tools for Pattern Recognition and Signal Analysis.  Over the last decade a number of theoretical arguments have been developed in order to justify their enhanced performance. The most widely known scenario is to look at them as maximum margin classifiers. Another approach is via learning theory arguments and the structural risk minimization principle, which leads to an optimal trade off between performance and complexity. An alternative path is to look at the cost function, associated with the SVMs, as a regularized minimizer that asymptotically tends to the Bayesian classifier. A less known viewpoint is the geometric one that leads to the notion of reduced convex hulls.  For the non-separable class case, the SVM solution is shown to be equivalent with computing the minimum distance between two reduced versions of the original convex hulls that “encircle” the two classes (for the two class case).

In this talk I will focus on the geometric approach and new results will be discussed concerning a) novel, necessary for our case, theorems concerning the structure and properties of the reduced convex hulls (RCH) and b) novel algorithms for computing the minimum distance between the resulting RCH´s. This problem is far from being trivial, since existing algorithms, which compute the minimum distance between convex hulls, rely on their respective extreme points. However, computing the extreme points of a reduced convex hull, as we have shown, is a computationally hard task of a combinatorial nature. A basic projection theorem, that we have shown, will be discussed that bypasses the combinatorial burden of the task and opens the way to employ geometric minimum distance algorithms to the SVM task.  Most important, this theorem “respects” inner products, thus allowing to the well known kernel trick to be easily incorporated into the algorithmic schemes, making them appropriate for the general nonlinear non-separable problem.

The derived geometric algorithms are much more efficient compared to the classical and widely used SMO algorithm and its versions. A number of tests with well known test beds have shown that, sometimes,   a gain of an order of magnitude in the number of kernel computations, for similar error rates, can be achieved. Furthermore, the new schemes are closer to our intuitive understanding of an iterative algorithm in simple geometric arguments.

16.00h: "A Weighted Kernel PCA framework for unsupervised learning"
Carlos Alzate (K.U. Leuven, ESAT-SCD)

Kernel PCA as a nonlinear generalization of PCA first maps the input data into some feature space via a nonlinear feature map induced by a kernel function and then performs linear PCA on the mapped data.  The objective is to find projected variables with maximal variance in this new feature space.  The solutions are given by the eigenvectors of the centered kernel matrix derived from the data and the corresponding eigenvalues indicate the amount of variance captured by the projection.  This technique has been widely used in the recent years for nonlinear feature extraction, density estimation and denoising.   Another unsupervised learning technique is spectral clustering which corresponds to a class of algorithms that make use of the eigenvectors of some data-driven matrix to group data points that are similar.  Spectral clustering algorithms are formulated as relaxations of graph partitioning problems that are generally NP-complete.  These relaxations take the form of eigenvalue problems involving a normalized affinity matrix containing pairwise similarities.  One drawback of spectral clustering is the fact that the clustering model is defined only for training data with no clear extension to new (out-of-sample) points.

In this talk, I will discuss a generalized formulation to kernel PCA based on the Least Squares Support Vector Machine (LS-SVM) formulation.  This formulation introduces weighting factors that can be used to change the L2 loss function associated to kernel PCA in order to achieve desirable properties such as robustness and sparseness in a fast and efficient way.  A different kind of weighting provides links to some spectral clustering methods with weighted kernel PCA as a unifying view.  Spectral clustering algorithms such as the normalized cut, the random walks model, the kernel alignment and the NJW algorithms are shown to be particular cases of weighted kernel PCA.  This unifying method allows to extend the clustering model to out-of-sample data points by using the projections onto the eigenvectors which becomes important for predictive purposes.  Simulations with toy examples and images in the context of denoising, clustering and segmentation will be presented.

Thu 15 - Sun 18 Feb-07 Mini-filmfestival "Man & Machine"
Cinema ZED, Leuven
In the framework of Kulturama  2007, Cinema ZED presents a mini filmfestival "Man & Machine" about cyborgs, robotics, artificial intelligence and transhumanism. More info in the attached pdf.
[PDF]

Thu 15 - Thu 15 Feb-07 EU workshop on Distributed/Hierarchical MPC
Esat 02.58
9:00 am-5:00 pm

website

Wed 14 - Wed 14 Feb-07 K. U. Leuven Seminars on Optimization in Engineering - Riccardo Scattolini and Joachim Ferreau
ESAT 00.62
3:00 pm-5:00 pm

Two talks on Automotive MPC, with Coffee break in between at 3:45 pm.

15 h: "Model Predictive Control in Automotive Systems"

Prof. Riccardo Scattolini
Dipartimento di Elettronica e Informazione
Politecnico di Milano

This talk presents some applications of Model Predictive Control (MPC) to automotive systems. Specifically, gasoline, Diesel and hybrid fuel cell vehicles are considered. In the first part of the talk, a detailed gasoline engine model is described and it is shown how to perform the optimal tuning of the engine maps by solving a suitable static optimization problem. Then, the control scheme is complemented with an MPC regulator to enhance the overall performances. Alternative strategies are also proposed by resorting to a “mixed” scheme where linear MPC is used together with a state feedback linearizing regulator and a nonlinear state observer. The second part of the talk is devoted to present the model of a turbocharged Diesel engine and to discuss its main control problems as well as the possible application of MPC for control of the air path. Finally, MPC is applied to the power flow management in an hybrid fuel cell vehicle.

*** Coffee Break ***

16h: "Model Predictive Engine Control using an Extended Online Active Set Strategy"

Hans Joachim Ferreau
University of Heidelberg and K. U. Leuven

In order to meet tight emission limits diesel engines are nowadays equipped with additional hardware components like an exhaust gas recirculation valve and a variable geometry turbocharger. Conventional engine control units use SISO control loops to regulate the corresponding actuators, although their effects are highly coupled. Moreover, these  actuators are subject to physical constraints which seems to make an advanced control approach like model predictive control (MPC) mandatory. In order to deal with MPC sampling times in the order of milliseconds, we employed an extension of the recently developed online active set strategy (OASES) for closed-loop control of a real-world Diesel engine at the Institute for Design and Control of Mechatronical Systems in Linz (Austria). The results to be presented show that predictive engine control based on online optimisation can be accomplished in real-time and leads to increased controller performance.

Slides

Tue 13 - Tue 13 Feb-07 K.U. Leuven Seminars on Optimization in Engineering - Christian Kirches
ESAT 00.62
3:00 pm-4:00 pm

"Robust Optimal Control of Discontinuous ODE Systems"

Christian Kirches,
University of Heidelberg

Abstract:
Optimal control problems based on systems modelled from a set of ODE or DAE equations allow for very efficient solution using the direct multiple-shooting approach. The underlying numerical methods will be briefly presented. The talk focuses on two numerical methods based on this approach. In the first part, a numerical method for the efficient treatment of implicit discontinuities (switches) in the ODE system is described. Challenges and new possibilities arise with the embedding of this approach in the direct multiple shooting method. The second part of the talk considers a linearization approach for the robustification of solutions obtained from direct optimal control problems. Finally, both areas of research will meet when an algorithm is presented that allows for robustifying discontinuous problems. All presented techniques will be motivated by and applied to a powertrain oscillation control problem by DaimlerChrysler AG of Stuttgart-Untertürkheim, Germany.

Slides

Thu 8 - Thu 8 Feb-07 K.U. Leuven Seminars on Optimization in Engineering - Philippe Toint
ESAT 02.58
4:00 pm
"Recursive trust-region methods for multilevel nonlinear optimization"

Philippe Toint (Univ. Namur)

Many large-scale optimization problems arise in the context of the
discretization of infinite dimensional applications.  In such cases, the
description of the finite-dimensional problem is not unique, but depends on
the discretization used, resulting in a natural multi-level description.
How can such a problem structure be exploited, in discretized problems or more
generally? The talk will focus on discussing this issue in the context of
unconstrained optimization and in relation with the classical multigrid
approach to elliptic systems of partial differential equations. Both
theoretical convergence properties of special purpose algorithms and their
numerical performances will be discussed.  Perspectives will also be given.

(in collaboration with S. Gratton, A. Sartenaer and M. Weber)

Thu 8 - Thu 8 Feb-07 K.U. Leuven Seminars on Optimization in Engineering - Sebastian Sager
ESAT 02.58
5:00 pm-6:00 pm
Direct Methods for Nonlinear Mixed--Integer Optimal Control Problems

Sebastian Sager (Univ. Heidelberg)

Many practical optimal control problems include discrete decisions. These
may be either time--independent parameters or time--dependent control
functions as gears or valves that can only take discrete values at any
given time. While great progress has been achieved in the solution of
optimization problems involving integer variables, in particular
mixed--integer linear programs, as well as in continuous optimal control
problems, the combination of the two is yet an open field of research.

We consider the question of lower bounds that can be obtained by a
relaxation of the integer requirements. For general nonlinear
mixed--integer programs such lower bounds typically suffer from a huge
integer gap. We convexify and relax the original problem and show that the
optimal solution of this continuous control problem yields the best lower
bound for the nonlinear integer problem.

Building on this theoretical result we present a novel algorithm to solve
mixed--integer optimal control problems, with a focus on discrete--valued
control functions. Our algorithm is based on the direct multiple shooting
method, an adaptive refinement of the underlying control discretization
grid and tailored integer methods. Its applicability is shown by
challenging applications, e.g., the energy optimal control of a subway
train with discrete gears and velocity limits, a batch process with time--
and tray--dependent reusage of slop (waste) cuts and port switching in
chromatographic separation processes.

Wed 31 - Wed 31 Jan-07 K.U. Leuven Seminars on Optimization in Engineering - Mario Milanese
ESAT 00.62
3:00 pm-4:00 pm
"FMPC: A Fast Implementation of Model Predictive Control, and its application to Semi-Active Suspension Control"

Prof. Mario Milanese, Politecnico di Torino, Italy

Model Predictive Control (MPC) is a model based control technique especially suited in dealing with input and/or state constraints. The online application of the procedure requires the solution of the an optimization problem at each sampling time, a task that can limit its use for systems with fast dynamics.

This motivates the recent research efforts devoted to develop computationally tractable MPC solutions. Indeed the control move at time results to be a continuous nonlinear static function of the system state. In [1], [2] explicit piece-wise linear solutions of the MPC problem have been introduced to compute the function in case of linear systems. They are based on a state space partition in polyhedral regions inside which the control law is an affine function of the system state that can be precomputed, stored, and implemented online. While such approach is quite attractive, as the online optimization is avoided, it may have serious limitations as, at each sampling time, the polyhedral region the initial state lies in, has to be determined. However, the number of subregions typically has a very fast increase with the dimension of state space and of the control horizon , leading to large computational complexity even for moderate values of state dimension and  control horizon.

A different approach has been introduced in [3] and [4], where a neural approximation of the static function is considered, based on the offline computation of the values of the function at a given number of states. The problems with such an approach are the trapping in local minima during the learning phase and the difficulty of handling the constraints in the image set of the function to be approximated. In the lecture, the Set Membership method proposed in [5] is presented for the approximation of the static function from the the offline computation of the values of the function at a given number of states. An approximating function is obtained fulfilling input and/or state constraint and whose computational time is independent on the MPC control horizon. The approach allows also to deal with Nonlinear Model Predictive Control (NMPC) problems. The effectiveness of the proposed methodology is shown by the application to a semi-active suspension control design [6].

References

[1] A. Bemporad, M. Morari, V. Dua, and E. N. Pistikopoulos, "The explicit
linear quadratic regulator for constrained systems," Automatica,
vol. 38, pp. 3-20, 2002.
[2] M. M. Seron, G. C. Goodwin, and J. A. De Dona, "Characterization
of receding horizon control for constrained linear systems," Asian J.
Control, vol. 5, no. 2, pp. 271-286, 2003.
[3] T. Parisini and R. Zoppoli, "A receding-horizon regulator for nonlinear
systems and a neural approximation," Automatica, vol. 31, no. 10, pp.
1443-1451, 1995.
[4] D. R. Ramirez, M. R. Arahal, and E. F. Camacho, "Min-max
predictive control of a heat exchanger using a neural network
solver," IEEE Trans. Contr. Syst. Technol., vol. 12, no. 5, pp.
776-786, Sep. 2004.
[5]. Canale and M. Milanese, "FMPC: A fast implementation of model
predictive control techniques," Proc. 16th IFACWorld
Congr., Prague, Czech Republic, 2005.
[6] M. Canale, M. Milanese, C. Novara, "Semi-active suspension control using Fast
Model Predictive Techniques", IEEE Transactions on Control Systems Technology,
vol. 14, no. 6, pp. 1034-1046, 2006.

Tue 30 - Tue 30 Jan-07 International Workshop on Modelling and Optimization of Power Generating Kites
ESAT 00.62
9:00 am-6:00 pm

# International Workshop on Modelling and Optimization of Power Generating Kites

## Leuven, January 30, 2007 9:00 a.m. - 6:00 p.m. Optimization in Engineering Center (OPTEC), K.U.Leuven, Room 00.62, ESAT-Building, Kasteelpark Arenberg 10, BE-3001 Leuven-Heverlee, Belgium

Aim of the one day workshop is to bring together experts and enthusiasts for a new class of large scale wind power generator based on tethered airfoils that has the potential to significantly increase the contribution of renewables to the global electricity supply. In particular, modelling, simulation, control and optimization methods shall be discussed, and first experiments with existing small scale prototypes shall be discussed. Two keynote lectures are given by W. Ockels (Delft Technical University) and M. Milanese (Politecnico di Torino).

The workshop is organized by Moritz Diehl, Boris Houska, and Andreas Ilzhoefer

Wed 17 - Wed 17 Jan-07 OPTEC Workshop - Nonlinear Model Predictive Control (NMPC) Tutorial Workshop
ESAT 00.57
1:30 pm-6:30 pm

Nonlinear Model Predictive Control (NMPC) Tutorial Workshop

Lecturers: Rolf Findeisen, Niels Haverbeke, Moritz Diehl

Course website

Aim of this 5 hour intensive workshop is to provide the participants with
knowledge about the basic concepts of nonlinear model predictive control
(NMPC). It is divided into five lectures, ranging from an introduction, over
stability theory, numerical solution, state estimation, and current research
topics. The workshop in particular profits from the presence of Rolf
Findeisen (Stuttgart University), an internationally recognized NMPC expert,
who visits OPTEC from Jan 16-18 and will give three of the five lectures. The
workshop can be quite interactive and participants are expected to ask many
nasty questions in the long coffee breaks ;-)

WORKSHOP PROGRAM AND ABSTRACTS:

13:30: Rolf Findeisen: Introduction and overview to NMPC (Slides)

The main focus in this lecture is laid on an introduction and historical
perspective of (nonlinear) predictive control. Specifically we outline the
basic principle of predictive control, reasons for the huge success of linear
inherent in NMPC.

14:15: Rolf Findeisen: Basic theory and stability of NMPC

Nonlinear model predictive control is based on the repeated solution of a
(finite) horizon open-loop optimal control problem subject to system dynamics
and input and state constraints. However, as is well known by now, optimality
does not automatically imply stability in the case of finite prediction
horizons. Different approaches to achieve closed-loop stability using finite
horizon lengths exist. The main purpose of this lecture is to review the
underlying main ideas and theoretical foundations for these approaches.

15:00: first coffee break

15:15: Moritz Diehl: Efficient Numerical Optimization for NMPC (Slides)

A necessary prerequisite for NMPC is the fast and reliable solution of
nonlinear optimal control problems in real-time. In the first part we give an
overview of general optimal control methods - dynamic programming, indirect
and direct approaches, with a focus on the last class. In the second part, we
outline crucial ideas behind real-time iteration algorithms, and present some
examples to illustrate their performance.

16:15: Niels Haverbeke: Introduction to Moving Horizon Estimation for
Nonlinear Systems (Slides)

The goal of state estimation is to reconstruct the state of a system from
process measurements and a model. In practice state estimators must address
many different challenges including nonlinear dynamics and hard constraints
on states or disturbances. The framework of moving horizon estimation (MHE)
is well suited to address these challenges and in many cases MHE gives
accuracy that is superior to the Extended Kalman filter which is widely used
in a large range of application areas. This talk aims at giving a general
introduction to MHE. We will highlight the key ingredients (stability,
smoothing, …) and show its practical use in a feedback control scheme (e.g.

17:00: second coffee break

17:30: Rolf Findeisen: Challenges and opportunities in predictive sampled data open-loop feedback control

In the first part we will discuss the issues of output feedback stabilization
using NMPC, since often the full state information as required for the
prediction is not available. For this purpuse we outline different
possibilities to achieve stabilizing output feedback control. In the second
part we focus on the question of control over networks subject to delays and
possible package losses. We will outline how these effects can be
counteracted and compensated, while guaranteeing stability.

18:30: End of the workshop

Wed 17 - Wed 17 Jan-07 K.U. Leuven Seminars on Optimization in Engineering - Rolf Findeisen
ESAT room 00.57
5:30 pm-6:15 pm
"Challenges and opportunities in predictive sampled data open-loop feedback control"

Rolf Findeisen (Univ. Stuttgart)

In the first part we will discuss the issues of output feedback stabilization
using NMPC, since often the full state information as required for the
prediction is not available. For this purpuse we outline different
possibilities to achieve stabilizing output feedback control. In the second
part we focus on the question of control over networks subject to delays and
possible package losses. We will outline how these effects can be
counteracted and compensated, while guaranteeing stability.

This talk is the last talk of the NMPC workshop taking place in the same
room, starting at 13:30 the same day. At 17:00 there will be a coffee
break to which all attendants are warmly invited.

Biographical Information:

Rolf Findeisen is C1-assistant professor at the Institute for Systems Theory
in Engineering at the University of Stuttgart. His main research areas are:
nonlinear model predictive control, output feedback control, optimization
based control and state estimation, differential algebraic systems, nonlinear
control, system theoretical methods in biomedical engineering and biological
systems; and the application of these methods in chemical, biological and
mechanical systems.

Selected publications:

* R. Findeisen, L.B. Biegler, and F. Allgower, editors. Assessment and Future
Directions of Nonlinear Model Predictive Control. Lecture Notes in Control
and Information Sciences. Springer-Verlag, Berlin, 2006.

* D. Mayne, S.V. Rakovic, R. Findeisen, and F. Allgower. Robust output
feedback model predictive control of constrained linear systems. Automatica,
1217-1222(42):7, 2006.

* M. Diehl, R. Findeisen, H.G. Bock, J.P. Schloeder, and F. Allgwer. Nominal
stability of the real-time iteration scheme for nonlinear model predictive
control. IEE Control Theory Appl., 152(3):296-308, 2005.

* R. Findeisen, L. Imsland, F. Allgwer, and B.A. Foss. Output feedback
stabilization for constrained systems with nonlinear model predictive
control. Int. J. of Robust and Nonlinear Control, 13(3-4):211-227, 2003.

Tue 9 - Tue 9 Jan-07 LERU Lecture - Bart De Moor
Auditorium Pentalfa at Leuven Gasthuisberg
4:00 pm-5:30 pm
"What engineers can do in Systems Biology and Bioinformatics"
Bart De Moor (K.U. Leuven, ESAT-SCD)

Tue 19 - Tue 19 Dec-06 K.U. Leuven Seminars on Optimization in Engineering - Ion Necoara
ESAT 01.57
11:00 am
"Robust control of a class of uncertain discrete event systems"

Ion Necoara (Delft University of Technology)

Abstract:
Discrete event systems are dynamical systems whose evolution equations changes in time by the occurrence of events. Discrete event systems that model only synchronization aspects are called max-plus-linear (MPL) systems. MPL systems can be described by models that are linear' in the max-plus algebra. In this presentation we derive solutions to three types of finite-horizon min-max control problems for uncertain MPL systems, depending on the nature of the control input over which we optimize: open-loop input sequences, disturbances feedback policies, and state feedback policies. We assume that the uncertainty lies in a polytope, and that the closed-loop input and state sequence should satisfy a given set of linear inequality constraints for all admissible disturbance realizations. Despite the fact that the controlled system is nonlinear, we provide sufficient conditions that allow to preserve convexity of the optimal value function and its domain. As a consequence, by employing recent results in polyhedral algebra and multi-parametric linear programming we prove that the min-max control problems can be either recast as a linear program or solved via N multi-parametric linear programs, where N is the prediction horizon. In some particular cases of the uncertainty description (e.g. interval matrices), using results from dynamic programming, we show that the min-max control problem can be recast as a deterministic optimal control problem.

Thu 7 - Thu 7 Dec-06 K.U. Leuven Seminars on Optimization in Engineering - Victor Zavala, Boris Houska
ESAT 00.62
2:00 pm-4:00 pm

14:00-14:45h "Simultaneous Nonlinear Programming Strategies for NMPC Applications"
Victor M Zavala (Carnegie Mellon University, Pittsburgh, USA)

Abstract- Nonlinear Model Predictive Control (NMPC) is an efficient framework for the control of large and constrained nonlinear systems. On the other hand, it involves the solution of complicated DAE-constrained optimization problems for which efficient solution strategies are required. While advances in numerical methods have enabled the successful implementation of challenging NMPC applications, new and more complex applications will inevitably require the extension of current optimization formulations, methods and algorithms and their adaptation to modern computer architectures.
In this talk we present a challenging industrial NMPC application: The High-Pressure LDPE Tubular Reactor Process. Motivated by the complexity of the system, a simultaneous collocation-based strategy has been applied for the solution of the associated DAE-constrained optimization problems. The approach discretizes the DAEs and solves the resulting large-scale nonlinear programming problems (NLPs). By doing so, efficient structure-exploiting linear algebra strategies can be implemented within a robust interior-point algorithm for the solution of the NLPs. The strategy has been used for the solution of large-scale parameter estimation and dynamic optimization problems. Finally, we present efficient real-time strategies for on-line implementations of NMPC and Moving Horizon Estimation tasks.

14:45-15:15h Coffee break

15:15-16:00h "Optimal Control of Power Producing Kites"
Boris Houska (University of Heidelberg)

In this talk we present optimization studies for kites that produce wind energy by periodically pulling a generator on the ground while flying fast in a crosswind direction. In the first part, we introduce the basic concepts of power generation with kites and compare them to conventional windmills. We also outline some ideas for large scale systems with many kites.
In the second part we show how to formulate an optimal control problem for a kite to obtain as much power as possible. We solve this nonlinear and unstable optimal control problem numerically with Bock's direct multiple shooting method and discuss the results. Finally, we give an outlook on Nonlinear Model Predictive Control (NMPC) methods for kites.

Wed 6 - Wed 6 Dec-06 Simon Stevin Lecture on Optimization in Engineering - Lorenz Biegler
Auditorium of the Arenberg Castle, Kasteelpark Arenberg 1, Heverlee Belgium
5:00 pm

First Simon Stevin Lecture on Optimization in Engineering

"Simultaneous Nonlinear Programming Strategies for Dynamic Optimization"
Lorenz T. Biegler (Chemical Engineering Department Carnegie Mellon University, Pittsburgh, USA)

pictures of Larry in Leuven

Abstract:
With the need to develop better designs and operating policies for dynamic nonlinear systems, it is important to consider efficient systematic strategies for the optimization of these systems. In particular, simultaneous approaches to dynamic optimization are favourable because of their efficiency and ability to handle complex dynamic features. This approach discretizes both the state and control profiles and solves a large-scale nonlinear program that results from the discretized system. Simultaneous optimization can be applied to optimize systems that are unstable, path constrained and high dimensional. Moreover, they are supported by two important elements:
- theoretical aspects of this approach have an equivalence to classical approaches such as Pontryagin's maximum principle
- efficient solution of large-scale nonlinear programs can be applied to these problems.
In addition to discussing these issues, a number of examples drawn from real-time optimization and data assimilation of water networks and chemical processes will be presented that demonstrate the benefits of this approach.

Bibliographical Information:
Lorenz T. (Larry) Biegler is currently the Bayer Professor of Chemical Engineering at Carnegie Mellon University, which he joined after receiving his PhD from the University of Wisconsin in 1981. His research interests are in the areas of computer aided process analysis and design and include flowsheet optimization, optimization of systems of differential and algebraic equations, reactor network synthesis and algorithms for constrained, nonlinear process control. Prof. Biegler has been a visiting scholar at Northwestern University, a scientist-in-residence at Argonne National Lab, a Distinguished Faculty Visitor at the University of Alberta, a Gambrinus Fellow at the University of Dortmund and a Fulbright Fellow at the University of Heidelberg. He has authored or co-authored over 200 archival publications, authored or edited seven books and presented numerous papers at national and international conferences.
He is the recipient of numerous awards including the AIChE McAfee Award (Pittsburgh Section), the AIChE Computers in Chemical Engineering Award, the ASEE Curtis McGraw Research Award and the Presidential Young Investigator Award from the National Science Foundation. He is a Fellow of the American Institute of Chemical Engineers and a member of SIAM, ACS and Sigma Xi. In addition, Professor Biegler has been an active consultant on process design and optimization strategies for the chemical and process industry.

The "Simon Stevin Lecture Series on Optimization in Engineering" is set up in order to promote optimization in engineering. For this aim, every quarter of the year an outstanding international scholar is invited to report on latest progress in the development of optimization algorithms and their applications in engineering.
Simon Stevin (1548-1620) was a Flemish mathematician and engineer. Among other, he helped to advance the use of decimal fractions, was the first to explain the tides by the attraction of the moon, and discovered the hydrostatic paradox. He made numerous inventions, among them a wind propelled carriage with sails, the "land yacht", which once impressed Prince Maurice of Orange as it moved faster than horses, in around 1600 on the beach between Scheveningen and Petten. Simon Stevin was fond of promoting the use of science in daily life and in craftmanship, and translated various mathematical terms into dutch. Among other, he introduced the dutch word for mathematics, "wiskunde".

Directly after the this winter's Simon Stevin Lecture, a little reception will be given at 6:30 pm in nearby Arenberg Castle, to which all attendants of the lecture are most warmly welcome!

***** REGISTRATION ENCOURAGED *****
Please send an e-mail with the subject "STEVIN" to Ida.Tassens@esat.kuleuven.be if you intend to participate in the event. No obligation, just to help us getting an idea how many people plan to come.

Mon 4 - Mon 4 Dec-06 SISTA Seminar - Athanasios Antoulas
ESAT 00.62
4:00 pm

"Model reduction from input/output measurements"
Thanos Antoulas (Rice University)

In many applications no explicit model of a given process is available. Instead input/output measurments of the process exist. The problem therefore consists in computing an appropriate approximate (reduced-order) model directly from these measurements. In this talk we will propose a method for achieving this objective. The main tool is the Loewner matrix. We will also show that this method can be extended to deal with passivity preserving model reduction.

Thu 30 - Thu 30 Nov-06 K.U. Leuven Seminars on Optimization in Engineering - Pierre-Antoine Absil
ESAT 00.62
4:00 pm

"Optimization on manifolds"
Pierre-Antoine Absil (UCL)

The world abounds with problems that can be formulated as finding an optimum of a real-valued cost function defined on a nonlinear search space that admits a differentiable manifold structure. When the nonlinear manifold is defined as a subset of a Euclidean space, the optimization-on-manifold approach can be thought of as solving an unconstrained optimization problem in a nonlinear space instead of an equality-constrained optimization problem in a linear space. To be competitive, this approach critically relies on the existence of numerically tractable local one-to-one mappings between the nonlinear manifold and a Euclidean space. Fortunately, several manifolds of great practical relevance admit such mappings.
In this talk, I will give an overview of optimization methods on manifolds and their applications, with an emphasis on the underlying geometric concepts and on the numerical efficiency of the algorithm implementations.

Mon 20 - Mon 20 Nov-06 SISTA Seminar - Karl Worthmann
ESAT 02.58
9:30 am-10:00 am
"High order approximations by sampled-data feedback"
Karl Worthmann (Mathematical Institute, University of Bayreuth, Germany)

A continuous time, nonlinear closed loop control affine system with a static state feedback u satisfying some control criteria is given. We consider the (zero order hold) sampled-data closed loop system for a fixed sampling period T>0 and a feedback u_T. Our goal consists in finding a suitable choice for u_T to reproduce as well as possible the continuous time system by the sampled-data system. More precisely, we are looking for necessary and sufficient conditions for determining whether an approximation of a given order is feasible or not. We assume that all functions involved in the continuous time system are sufficiently smooth and that the control of the discrete time system is locally bounded. We consider an approach - based on a comparision between the Taylor series expansion of the continuous time and the Fliess series expansion of the sampled-data system - that uses geometric control theory. We are in particular interested in asymptotic estimates, i.e. the system's behaviour for T tending to zero. This delivers control laws ensuring an approximation of a given order p after one sampling interval and consequently of order p-1 for all t=iT in arbitrary compact time intervals [0,t_1]. We derive necessary and sufficient conditions for the existence of suitable sampled-data feedback laws for the desired order and - in case of existence - provide explicit formulas and algorithms for these controllers. The results will be illustrated with two examples.
We indicate the restrictions of the described technique for the larger sampling periods present in practical applications. The mentioned problems may be solved via model predictive control or iterative optimisation methods. We present the latter by means of the Moore Greitzer jet engine example.

Thu 16 - Thu 16 Nov-06 K.U. Leuven Seminars on Optimization in Engineering - Jan Albersmeyer, Niels Haverbeke
ESAT 01.60
4:00 pm

"Automatic Differentiation and Efficient Sensitivity Generation for DAE Systems"
Jan Albersmeyer (Interdisciplinary Center for Scientific Computing, University of Heidelberg)

Many modern optimization algorithms rely on accurate derivative information for the objective and the constraints. Also in system analysis and for model reduction problems derivatives play an important role. In this talk we will give a short introduction to Automatic Differentiation (AD), an easy-to-use approach to obtain derivatives of functions very efficiently with nearly machine precision. There exist mainly two "modes" of AD, which both will be explained shortly.
In the second part we will present the prinicple of "Internal Numerical Differentiation", which we use to obtain efficiently derivatives of solutions of initial value problems with respect to initial values, parameters and controls, the so-called sensitivities. We show how this idea can be used inside an adaptive BDF-Integrator (DAESOL-II) to calculate sensitivities for initial value problems of semi-implicit DAEs of index 1. We will present the "forward" mode, which can be used to calculate directional sensitivities, and the newly implemented "adjoint" mode, which allows for efficient calculation of gradient information.

"The good and the bad of Moving Horizon state Estimation"
Niels Haverbeke (K.U. Leuven, ESAT-SCD)

The goal of state estimation is to reconstruct the state of a system from process measurements and a model. In practice state estimators must address many different challenges including nonlinear dynamics and hard constraints on states or disturbances. The framework of moving horizon estimation (MHE) is well suited to address these challenges and in many cases MHE gives accuracy that is superior to the Extended Kalman filter which is widely used in a large range of application areas. This talk aims at giving a general introduction to MHE. We will highlight the key ingredients (stability, smoothing, …) and show its practical use in a feedback control scheme (e.g. in cascade with MPC). The talk will conclude with simulation results of nonlinear MHE based model predictive control applied to a biomedical problem.

Thu 26 - Thu 26 Oct-06 K.U. Leuven Seminars on Optimization in Engineering - Juan Camino
ESAT 01.60
4:00 pm

"Solving Matrix Inequalities whose Unknowns are Matrices"
Juan Camino (K.U. Leuven, PMA)

This talk presents algorithms for numerical solution of convex matrix inequalities in which the variables naturally appear as matrices. This includes, for instance, many systems and control problems. To use these algorithms, no knowledge of linear matrix inequalities (LMIs) is required. However, as tools, they preserve many advantages of the linear matrix inequality framework. Our method has two components: 1) a numerical algorithm that solves a large class of matrix optimization problems; 2) a symbolic Convexity Checker'' that automatically provides a region which, if convex, guarantees that the solution from (1) is a global optimum on that region. The algorithms are partly numerical and partly symbolic and since they aim at exploiting the matrix structure of the unknowns, the symbolic part requires the development of new computer techniques for treating noncommutative algebra.

Tue 24 - Tue 24 Oct-06 Study day IAP V/22 - Large Graphs and Networks
Auditorium 02.56, College De Valk, Building DVI, Tiensestraat 41, 3000 Leuven
9:00 am-5:00 pm

Mon 23 - Fri 27 Oct-06 De Vlaamse Wetenschapsweek
Dept. Elektrotechniek
Tijdens de Vlaamse Wetenschapsweek worden er overal in Vlaanderen allerlei activiteiten rond wetenschap en technologie georganiseerd. Onder het motto 'Wetenschap in de kijker' doen zo'n 25.000 leerlingen van de 3de en 4de graad secundair onderwijs aan wetenschappelijk onderzoek in universiteiten, hogescholen en wetenschappelijke instellingen. Ook musea, wetenschappelijke verenigingen, sterrenwachten en bibliotheken zetten die week speciale activiteiten op het getouw. De Vlaamse Wetenschapsweek wordt in de even jaren in oktober georganiseerd. De volgende Wetenschapsweek zal plaatshebben in oktober 2006.

Thu 19 - Thu 19 Oct-06 SISTA Seminar - Frizo Janssens, Raf van de Plas
ESAT 01.60
4:00 pm

"Bioinformatics: a hybrid textual and bibliometric analysis (1981-2004)"
Frizo Janssens (K.U. Leuven, ESAT-SCD)

We analyze a set of about 7500 bioinformatics-related papers from the ISI Web of Science database, publication years 1981-2004. For delineating this complex inter-disciplinary field, the unconditional criteria of the bibliometric retrieval strategy included 4 core journals and a title keyword search. The resulting set of papers was further extended with publications that at least three times cite or are cited by the core. To retrieve associated MeSH terms, each included paper was also matched against Medline. The complete document set is examined using text mining and bibliometric techniques and we consider author, institutional and international collaboration networks. Given that the performance of clustering and classification of scientific papers can significantly be improved by integrating textual contents with the structure of the citation graph, we proceed with a hybrid clustering method to map the cognitive structure of the field. An optimal clustering with 9 clusters is described in quantitative and qualitative sense and various term networks based on different vocabularies provide diverse views on the same clustering outcome. Next, a dynamic clustering is obtained by matching and tracking of clusters that are found in consecutive periods defined by time windows on the set.

"Exploration of Biochemical Tissue Composition via Imaging Mass Spectrometry guided by Multivariate Data Analysis"
Raf Van de Plas (K.U. Leuven, ESAT-SCD)

MALDI-based Imaging Mass Spectrometry (IMS) is an analytical technique that provides the opportunity to study the spatial distribution of biomolecules including proteins and peptides in organic tissue. IMS measures a large collection of mass spectra spread out over an organic tissue section and retains the absolute spatial location of these measurements for analysis and imaging. The classical approach to IMS imaging, producing univariate ion images, is not well suited as a first step in a prospective study where no a priori molecular target mass can be formulated. The main reasons for this are the size and the multivariate nature of IMS data. In this talk we describe the use of multivariate approaches, such as principal component analysis and k-means clustering, as a prospective pre-analysis tool, to identify the major spatial and mass-related trends in the data and to guide further analysis downstream. First, a conceptual overview of the multivariate methods for IMS is given. Then, we demonstrate these approaches on an IMS data set collected from a transversal section of the spinal cord of a standard control rat.

Thu 12 - Thu 12 Oct-06 K.U. Leuven Seminars on Optimization in Engineering - Moritz Diehl, Hans Joachim Ferreau, Matthieu Guilbert
ESAT 00.62
2:30 pm-5:30 pm

14.30-15.30h: "Dynamic Optimization in Engineering - Algorithms and Applications"
Prof. Dr. Moritz Diehl (ESAT K.U.Leuven, COE Optimization in Engineering)

Aim of the talk is to illustrate how challenging application problems in engineering can solved by help of dynamic optimization, i.e., solution of optimization problems with underlying dynamic system models in form of ordinary or partial differential equations. In order to exploit the full potential of dynamic optimization in engineering applications, specialized algorithms and approximation methods are needed. We argue that the best methods are very different from the "black-box approach" that just uses a good ODE/PDE simulation routine within good optimization routine. If both simulation and optimization are carried out together in all-at-once approaches, often orders of magnitude in efficiency can be realized and previously untreatable problems become solvable.
We will present three recent applications along with the underlying dynamic optimization algorithms:
a) robust open-loop control of an exothermic chemical reactor that shall avoid runaways (including experiments)
b) nonlinear model predictive control (NMPC) of a distillation column (including experiments)
c) periodic optimal control of tethered airfoils for a novel way of large scale wind power generation (so far only computer simulations)

15.30-16.00h: coffee break

16.00-16.45h: "An Online Active Set Strategy for Fast Parametric Quadratic Programming and Application to Predictive Engine Control"
Hans Joachim Ferreau (Univ. Heidelberg)

Nearly all algorithms for model predictive control (MPC) rely on solving convex quadratic programs in real-time. In this talk, we develop a specially tailored online active set strategy for the fast solution of parametric quadratic programs arising in MPC. Our strategy exploits solution information of the previous quadratic program (QP) under the assumption that the set of active constraints does not change much from one QP to the next. Furthermore, we present a modification where the CPU time is limited in order to make it suitable for strict real-time applications. An efficient implementation of the proposed online active set strategy is described and its performance is demonstrated with two challenging test examples. One of these was designed for controlling a real-world Diesel engine with sampling times of a few milliseconds. In these examples, our strategy turns out to be an order of magnitude faster than a standard active set QP solver.

16.45-17.30h: "Optimization of industrial robot trajectories subject to real physical limitations"
Matthieu Guilbert (INRIA, Grenoble)

The optimization of robot trajectories usually considers bounds on motor accelerations, velocities or torques, but that does not really reflect the real physical limitations of a robot. We will therefore focus here on the optimization of robot trajectories subject to real physical limitations such as temperature limitations. We also want to point out that industrial robots are often integrated in robotic cells which are usually difficult or even impossible to model. We propose therefore to decompose the optimization into two levels: the first algorithm is based on models and a discretization of the velocity profile, and the second one is based on velocity and torque measures on the robot and on derivative free algorithm. The proposed algorithm gives good numerical and experimental results on complex real world robotic applications.

Thu 5 - Thu 5 Oct-06 SISTA Seminar - G. Skordev
ESAT 01.60
4:00 pm
"Old and new about strict self-similar sets : constructions and dimensions"
G. Skordev, Center for Complex systems and Visualization, University of Bremen

Abstract: Deterministic and random dynamical constructions of strict self-similar sets such as Moran substitutions, Hutchinson's iterated function systems, Mauldin-Williams graph directed constructions, 2-D substitutions and automatic sequences are discussed. There are several "dimensions" that charactarize self-similar sets: self-similar - , box-counting - and Hausdorff dimensions. We discuss these dimensions and the relations between them for strict self-similar sets.

Mon 21 - Wed 23 Aug-06 4th International Workshop on Total Least Squares and Errors-in-Variables Modeling
Arenberg castle
This interdisciplinary workshop is a continuation of 3 previous workshops which were held in Leuven, Belgium, August 1991, 1996, and 2001 and aims to bring together numerical analysts, statisticians, engineers, economists, chemists, etc. in order to discuss recent advances in Total Least Squares techniques and errors-in-variables modeling.

Tue 4 - Tue 4 Jul-06 One-day Workshop on "Synchronization in Complex Networks"
ESAT 00.62

One-day Workshop on "Synchronization in Complex Networks"

Date: Tuesday July 4, 2006
Place: Department of Electrical Engineering, ESAT 00.62, K.U. Leuven, Belgium

-Preliminary programme

09.20 Welcome - Johan Suykens (K.U.Leuven, ESAT-SCD)

09.30-10.00: The hyperbolic Plykin attractor can exist in neuron models - Vladimir Belykh (Volga State Academy, Nizhny Novgorod)
In this talk we present a construction of three-dimensional vector field which is related to a neuron model and can generate the hyperbolic Plykin attractor. The contents of the talk is as follows: Strange hyperbolic attractors and the Plykin example; The bursting neuron model; Global section construction and the Poincare return map; Multi-loop homoclinic bifurcations and Plykin-like attractors.

10.00-10.30: Synchronous regimes in ensembles of locally diffusively coupled oscillators - Grigory Osipov (Nizhny Novgorod State University)
We present different aspects of synchronization in chains and lattices of locally interconnected nonidentical oscillatory elements. Main kinds of collective behavior are discussed for prototypical continuous and discrete in time systems.

10.30-10.45: Break

10.45-11.15: Learning partial synchronization regimes - Daniel Hillier (P. Pazmany Univ. Budapest)
We demonstrate that using a numerical optimization framework it is possible to explore and analyze the information processing capabilities of non-linear oscillator networks. It will be shown how a properly formulated cost function can be used to achieve synchronization and at the same time impose qualitative behavior on an array of chaotic oscillators. Our approach aims to add an application motivated aspect to existing results that so far focused on conditions for synchronization. To demonstrate the potential of the optimization framework we show on a simple case study that network configurations corresponding to partial synchronization regimes can be learned. This can also be done with imposing various qualitative behavior on individual oscillators.

11.15-11.45: Learning spatio-temporal phenomena on cellular neural networks chips - Samuel Xavier-de-Souza (K.U.Leuven, ESAT-SCD)
The problem of learning spatiotemporal behavior with cellular neural networks is discussed in this talk. Its link with trajectory learning for recurrent neural networks is analyzed. Despite of similarities, the two learning problems have underling differences which makes a direct mapping non-trivial. An optimization methodology that also incorporates time instants as learning parameters is presented. This method imposes only the ordering of the learning points and relax their time schedule. Results of simulation and on-chip learning are presented for different classes of spatiotemporal phenomena.

11.45-12.00: Variety of synchronous regimes in ensembles of neuron-like oscillators - Liubov Averyanova (Nizhny Novgorod State University)
We study synchronous behavior in ensembles of locally diffusively coupled (i) nonidentical Bonhoeffer - van der Pol oscillators, and (ii) nonidentical Huber-Braun oscillators. We show that in a chain of N elements 2^{N-1} different regimes of global synchronization can coexist.

12.00-14.00: Lunch

14.00-14.30: Complete synchronization in diffusively coupled Morris-Lecar neuronal systems - Evgeniya Pankratova (Volga State Academy, Nizhny Novgorod)
In the present talk a neuronal network, consisting of elements whose dynamics is described by the system of Morris-Lecar differential equations, is considered. The study of complete synchronization in networks with different coupling configurations is performed in the framework of connection graph stability method. This recently developed method allows to calculate upper bounds for global synchronization in complex networks of mutually coupled dissipative oscillators. In the present talk the regime of complete synchronization for a chain of Morris-Lecar systems and an ensemble that simulates the behavior of a pair diffusively coupled pace-maker nerve cells is examined. In the framework of connection graph stability method the change of the synchronization threshold due to the both growth of the distance between the pace-maker neurons and increase of elements, that are directly coupled with the pace-maker neuron, is analyzed.

14.30-15.00: Global optimization with coupled local minimizers excited by Gaussian white noise - Serkan Gunel (Univ. Izmir)
Stochastic Coupled Local Minimizers (SCLMs) are presented for global optimization of smooth cost functions. This extends the deterministic coupled local minimizers (CLMs) whose best performing elements impose their dynamics to others by means of master-slave synchronization. The elements of the CLM network are excited by Gaussian white noise. An adaptive cooling schedule that effectively decouples the noise sources from the network when a solution candidate is agreed among all CLMs is proposed.

15.00-15.15: Break

15.15-15.45: Fluctuational phenomena in Josephson electronic devices - Andrey Pankratov (Institute for Physics of Microstructures, Russian Academy of Sciences, Nizhny Novgorod)
The aim of this presentation is to give a short review of recent achievements in the theory of fluctuational phenomena in Josephson junctions. In particular, the effect of resonant activation and noise suppression in short Josephson junctions subjected to periodic and pulse driving will be given. Peculiarities of complex fluctuational dynamics in microwave hysteretic SQUID will be outlined. Noise-induced dynamics of long Josephson junctions, studied by direct computer simulations of sine-Gordon equation with noise, will be described.

15.45-16.15: Cooperative behaviour in coupled simulated annealing processes - Samuel Xavier-de-Souza (K.U.Leuven, ESAT-SCD)
In this talk, we present a strategy for solving global optimization problems based on the information exchanged within multiple Simulated Annealing processes. Cooperative behavior emerges among these processes by coupling their acceptance probability functions. This approach does not only outperforms Parallel Simulated Annealing, but also reduces the sensitivity to initialization by means of a variance control of the acceptance probabilities.

16.15-17.00: Poster session

Poster session:
Participants who are interested in presenting a poster are asked to contact Johan.Suykens@esat.kuleuven.be

Lunch:

Route description:
http://www.esat.kuleuven.ac.be/info/route.en.php

Tue 27 - Tue 27 Jun-06 Doctoral Presentation - Sven Maerivoet
Auditorium Wolfspoort (Huis Bethlehem), Schapenstraat 34, Leuven
5:00 pm

"Modelleren van Verkeer op Autosnelwegen: State-of-the-Art, Numerieke Data Analyse, en Dynamische Verkeerstoedeling"
Sven Maerivoet (K.U. Leuven, ESAT-SCD)

Terwijl de filevorming in steden en landen een immer-toenemende trend vertoont, wordt het modelleren van wegverkeer een steeds maar actiever vakgebied. Daar waar reeds vele inspanningen werden gedaan met betrekking tot de lokale en globale regeling van verkeersstromen, is ons onderzoek gericht op het modelleren van wegverkeer op autosnelwegen. Het doel van ons onderzoek is drievoudig; eerst geven we een volledige standaard omtrent nomenclatuur binnen het gebied van de verkeerskunde, gebaseerd op een consistente verzameling notaties. Dit wordt gevolgd door een gedetailleerd literatuuroverzicht omtrent de wiskundige modellen die gebruikt worden om het verkeer op wegen te beschrijven, dit vanuit zowel het standpunt van transportplanning als stromingsmodellen. Speciale aandacht gaat uit naar de klasse van cellulaire automaatmodellen van wegverkeer. Ten tweede voeren we een verkennende data analyse van ruwe verkeersmetingen uit, waarbij we de operationele karakteristieken van enkelvoudige lusdetectoren bespreken. Verder reiken we onderzoekers middelen aan om statistische uitschieters op te sporen, om op een snelle manier structurele en incidentele storingen van detectors te beoordelen, om reistijden te schatten op een off-line manier, gebaseerd op ruwe cumulatieve tellingen, en om een visuele voorstelling van de dynamica van verkeersstromen in tijd en ruimte te verkrijgen. Tot slot, voorzien we, binnen de context van simulatie-gebaseerde dynamische verkeerstoedeling, een duidelijke methode om zowel de problemen van de keuzes van vertrektijdstip en route op sequentiële wijze te combineren, dit gebouwd rond een verkeersstroommodel dat uitgewerkt wordt als een computationeel efficiënte cellulaire automaat. Met betrekking tot de literatuur onderscheiden onze bijdragen zich doordat ze een synthese vormen van de benaderingen voor het beschrijven van wegverkeer, terwijl dergelijke samenvattingen tot op heden enkel verspreid bestonden. Om een globaal beeld te krijgen met betrekking tot de kwaliteit van verkeersmetingen, bieden wij daarnaast methodes aan die kunnen omgaan met grootschalige data, dit in tegenstelling tot het meeste onderzoek naar de numerieke analyse van verkeersmetingen wat vaak slechts op beperkte data wordt uitgevoerd. Tenslotte met betrekking tot de vele benaderingen van het paradigma van simulatie-gebaseerde dynamische verkeerstoedeling, stellen wij een methodologie voor die de keuze van het vertrektijdstip sequentieel met de routekeuze integreert.

Promotoren: Bart De Moor, Ben Immers

Wed 7 - Wed 7 Jun-06 Doctoral Presentation - Marcelo Espinoza
Auditorium of the Arenberg Castle
5:00 pm
"Structured Kernel Based Modeling and its Application to Electric Load Forecasting"
Marcelo Espinoza (K.U. Leuven, ESAT-SCD)

In the nonlinear system identification and forecasting of time-series, important challenges are related to the accurate modeling by incorporation of prior knowledge and the estimation of such models from large scale datasets. In this thesis, the main scope is structured kernel based modeling and its application to electric load forecasting. We take as a starting point the Least-Squares Support Vector Machines (LS-SVM) formulations for nonlinear regression. The primal-dual optimization framework can be extended to incorporate structured elements available from prior knowledge about the problem. The results are derived for the case of imposing symmetry to the estimated nonlinear model, imposing an additional parametric term for a new set of regressors and incorporating autocorrelation in the noise process of the regression. For each of these extensions, the goal is to include the additional structures in the form of equality constraints such that the resulting problem or subproblem remains convex, and Mercer's theorem can be applied with the use of a positive definite kernel and a kernel induced feature map. The prior information contained in the additional constraints becomes embedded at the kernel level, such that it can be used directly to evaluate the models at new datapoints. This property makes a contribution in terms of modularity of the model formulation, in the sense that different types of prior knowledge can be tested in practice simply by changing the kernel function being used. Furthermore, large scale versions of the different LS-SVM extensions can be formulated in primal space by using the Nystrom method (which delivers finite dimensional approximations to the feature map as shown in the area of Gaussian processes) in the same way as for original fixed-size LS-SVMs. By considering each of the developed extensions as building blocks, a modular framework for the case of nonlinear system identification is further proposed. It is shown that this framework can be used for the estimation of NARX and AR-NARX model structures, with different possible parameterization, exploiting the practical advantage of formulating the model in dual space and estimation in primal space for large sample sizes. The nonlinear system identification methods have been tested in a real-life industrial application by considering the short-term electricity load forecasting problem. Comparing different structures, we find that nonlinear models can capture the behavior of the load series and generate more accurate forecasts than the linear models, particularly when comparing not only black-box structures but also more structured representations. It is shown that the modular approach proposed in this thesis can be quite successful in the definition, estimation and final forecasting performance of nonlinear time series models.

Promotors: Bart De Moor, Ronnie Belmans

Wed 3 - Wed 3 May-06 Math-ESAT-CW Seminar - Nick Trefethen
ESAT Aud B
4:00 pm-5:00 pm
"Computed eigenmodes of planar regions"
Nick Trefethen (Oxford University)

Recently developed numerical methods make possible the high-accuracy computation of eigenmodes of the Laplacian for a variety of "drums" in two dimensions, or as some physicists prefer to call them, problems of "quantum billiards". A number of computed examples will be presented together with a discussion of their implications concerning bound and continuum states, symmetry and degeneracy, eigenvalue avoidance, resonance, localization, eigenvalue optimization, perturbation of eigenvalues and eigenvectors, and the problem of "can one hear the shape of a drum?".

Tue 25 - Tue 25 Apr-06 SISTA Seminar - Mihaly Petreczky (CWI, Amsterdam)
Aud. A
2:00 pm-3:00 pm
In the talk I will give an overview of realization theory for a number of classes of hybrid systems. Hybrid systems are control systems which contain both discrete and continuous components. In this talk the realization problem, that is, the problem of finding a minimal state-space model from the input-output behavior, will be investigated. I will mostly speak about hybrid systems without guards, that is, hybrid systems such that the change of discrete states is independent of the continuous states. I will also assume that the continious dynamics is specified by linear or bilinear differential equations. For such systems I will present characterizations of minimal realizations and necessary and sufficient conditions for existence and uniqueness of a realization. I will discuss in more detail the algorithmic aspects of realization theory. That is, I will present algorithms for checking minimality, computing minimal realizations. I will also present algorithms for computing minimal realizations from input-output data. A matrix factorization algorithm for the generalized Hankel-matrix lies in the core of most of these algorithms. This algorithm is very similar to the one for quasi-realization of Hidden Markov models and for realization of bilinear systems.

Thu 23 - Thu 23 Mar-06 SISTA Seminar - Dirk Aeyels - Samuel Xavier-de-Souza
ESAT 00.62
4:00 pm

"A dynamical model for cluster formation"
Dirk Aeyels (University of Gent, SYSTEMS)

We present a dynamical model of mutually attracting agents. The long term behavior is characterized by agents organized into several clusters; transitions to new cluster configurations take place, depending on the intensity of the attraction. The number of clusters and which agents inhabit which clusters is determined by a set of inequalities in the parameters of the model. We consider the case where all agents are mutually attracted. The analysis of the clustering process is supported by a mathematical proof. As an illustration of the scope of the model, we indicate and discuss some applications.

"Cooperative Behaviour in Coupled Simulated Annealing Processes"
Samuel Xavier-de-Souza (K.U. Leuven, ESAT-SCD)

Suppose you have a task. You need to find the deepest valley in a vast and hilly landscape, but you are not alone. Together with you, there are a dozen of explorers, all equipped with a altimeter and a radio communicator. What would you do? which search strategy would you define? If you were alone, you would be facing a typical global optimization problem. However, you have the other explorers in your advantage. Cooperation is clearly necessary in order to perform the task efficiently. In this seminar, we present a strategy to solve your problem based on the information your co-workers are exchanging. In order to solve global optimization problems with many local optima, we force cooperative behaviour to emerge among Simulated Annealing processes by coupling their acceptance probability functions. This approach does not only outperforms Parallel Simulated Annealing, but also reduces the sensitivity to initialization by means of a variance control.

Thu 9 - Thu 9 Mar-06 K.U. Leuven seminars on Optimization in Engineering - Yurii Nesterov
ESAT 00.62
4:00 pm

"Gradient methods for convex optimization in relative scale"
Yurii Nesterov, CORE, Universit'e Catholique de Louvain (UCL).

In this talk we propose new efficient gradient schemes, which can solve a structural convex optimization problem with certain {em relative} accuracy $delta$. As an example, we consider two non-trivial classes of linear programming problems, which can be solved in $O({sqrt{n ln m} over delta}ln n)$ iterations of a gradient-type method ($n$ and $m$, $n < m$, are the sizes of the corresponding problems). The proposed schemes are based on preliminary computation of an ellipsoidal rounding for some polytopes in $R^n$. In both cases this computation can be performed very efficiently, in $O(n^2m ln m)$ operations at most.

Thu 26 - Thu 26 Jan-06 K.U. Leuven seminars on Optimization in Engineering - Kristiaan Pelckmans
ESAT 00.62
4:00 pm

"On the Regularization Path in Bi-criterion Optimization Problems"
Kristiaan Pelckmans (K.U. Leuven, ESAT-SCD)

In this seminar, we review and discuss the use of the so-called regularization path in bi-criterion optimization problems. In short, the regularization path characterizes the optimal solutions for varying scalarizations of the two competing objectives. Theoretical concepts, as well as practical algorithms for computing the path are reviewed, and the discussion is related to paradigms as simplectical methods.
This concept has found important applications towards the design of algorithms for machine learning, as confirmed by a successful workshop on NIPS 2005. Here the importance of computing the full path was emphasized beyond its mere computational advantage. A classical application is found in the 1-norm regularized least squares estimator (LASSO), but this talk will also discuss recent extensions towards clustering techniques as well as its relevance for automatic model selection procedures. The presentation examplifies those concepts from a perspective of system identification.

Wed 1 - Wed 1 Jun-05 Doctoral Presentation - Luc Hoegaerts
Auditorium of the Arenberg Castle
5:00 pm
"Eigenspace Methods and Subset Selection in Kernel based Learning"

Luc Hoegaerts (K.U. Leuven, ESAT-SCD)

The last decades have been characterized by a considerable increase in the amount and complexity of information, far exceeding the ability to process and interpret the data. Inference tools for data mining and knowledge extraction draw upon techniques from artificial intelligence, machine learning and statistics. In this thesis we generalize or extend some of those methods that are based on the kernel paradigm-the latest generation of models that have the capability to learn (non)linear relations from a given collection of pairwise examples, like for example in classification or regression problems. We investigated upon five main topics: (i) we derive a primal-dual interpretation for Kernel Partial Least Squares and provide a unifying framework for a set of existing least squares kernel based regression methods in a Reproducing Kernel Hilbert space, (ii) we endow these methods with a sparse formulation via the Nystrom approximation, which permits to perform kernel regression on very large data sets with limited loss of accuracy, (iii) we explore possible up/downdating strategies for Least Squares Support Vector Machines, with induced data point selection criteria, (iv) we devise a novel tracking algorithm to efficiently approximate the dominant eigenspace of the kernel matrix and (v) we derive a generalized Grassmann-Rayleigh Quotient Iteration for the computation of the best rank-(R1,R2,R3) approximation of higher-order tensors.

Promotors: Prof. Joos Vandewalle, Prof. Johan Suykens

Tue 31 - Tue 31 May-05 Doctoral Presentation - Kristiaan Pelckmans
Auditorium of the Arenberg Castle (changed!)
5:00 pm
"Primal-dual kernel machines"

Kristiaan Pelckmans (K.U. Leuven, ESAT-SCD)

This doctoral dissertation gives an overview of new advances obtained in machine learning using primal-dual kernel machines. The general objective is the formulation and study of a broad methodology assisting the user in making decisions and predictions based on collections of observations. The proposed approaches are mainly studied in a context of convex optimization, the main contribution of the research is found in the formulation of an hierarchical programming problem which is employed to automate the problem of model selection and the formulation of hierarchical kernel machines. Additional results include the formulation of extensions to the method of primal-dual kernel machines for handling new tasks, various ways for dealing with model complexity control and regularization, estimation of semi-parametric models and additive models using componentwise kernel machines, the handling of missing values and censored output observations, and others. The bottomline is to illustrate the use of the primal-dual argument as a powerful building block for the design and analysis of algorithms.

Promotors: Prof. Bart De Moor, Prof. Johan Suykens

Thu 3 - Thu 3 Mar-05 SCORES Seminar - Marcelo Espinoza - Geert Deconinck
ESAT Aud A
4:00 pm-5:30 pm
"Using a priori information for model structures within the LS-SVM framework"
Marcelo Espinoza (K.U.Leuven, ESAT-SCD)

"From electrical energy to computer architectures: ELECTA research overview"
Geert Deconinck (K.U.Leuven, ESAT-ELECTA)

Thu 10 - Thu 10 Apr-03 SISTA Seminar - Ben Immers - Sven Maerivoet - Wouter Favoreel
ESAT 00.62
4:00 pm-6:00 pm
"A Survey on the Qualitative Behaviour of Traffic Cellular Automata"
Sven Maerivoet (K.U.Leuven, ESAT-SCD)

Considering the framework of microscopic road traffic flow models, we investigate those variants that are based on the statistical physics' cellular automata programming paradigm as a discrete dynamical system. These kind of models are very efficient from a computational point of view, as they use simple rulesets to describe vehicle movements on a rather coarse scale, allowing easy parallellisation through distributed computing. Our study elaborates on some of the qualitative effects that emerge from the dynamical processes behind these systems of traffic flows, involving the behaviour of vehicles in the time and space evolution of a cellular automaton. With respect to this, we give a retrospective on several classic TCA models, as well as some of their advantages and deficiencies.

"Video detection for traffic applications"
Wouter Favoreel (Traficon)

In this presentation an overview of the Traficon video detection technology will be given. Special emphasis on the three main application areas will be given: tunnels, highways and intersections. In a second part of the talk an overview of the new developments will be presented. Finally a state estimation and control application will be presented for a highway network.

Thu 13 - Thu 13 Mar-03 SISTA Seminar - Dan Sorensen - Luc Hoegaerts - Tijl Debie
ESAT 00.62
4:00 pm-6:00 pm
4:00pm: "Projection Methods for Large Eigenvalue Problems: Analysis, Extensions, and Applications"
Dan Sorensen (Rice University & visiting professor at K.U.Leuven ESAT-SCD-SISTA)

This talk will discuss developments during the past decade of projection methods for large eigenvalue problems. During this period very effective algorithms and software have appeared, and considerable advances in the ability to deal with non-symmetric matrices have resulted. Krylov methods are a primary example and developments in understanding and implementing restarting have enabled the construction of effective and robust software based upon the Arnoldi process. ARPACK is one example. This software has been successfully applied to linear stability problems in CFD on the order of 16 million variables.
Recently, a convergence analysis has emerged that adequately explains the behavior of Krylov methods with polynomial restarting. Moreover, new techniques for accelerating convergence without a direct matrix factorization have been developed to extend the domain of applications that can be effectively solved with such software.
Finally, large applications in various areas of scientific computing are now being solved by posing problems in the form of selected invariant subspace computation. This is a problem ideally suited for restarted Krylov methods and existing software can be used. Examples including regularization methods for large least squares problems, solution of Sylvester equations, and model order reduction for large state space control systems will be presented. There will be specific focus on regularization during this talk.

5:00pm: "Subspace regression in reproducing kernel Hilbert space"
Luc Hoegaerts (K.U.Leuven ESAT-SCD-SISTA)

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 primal-dual optimization problem formulation. The methods are illustrated and compared on a number of examples.

5:30pm: "Canonical correlation analysis, a different view and applications"
Tijl De Bie (K.U.Leuven ESAT-SCD-SISTA)

Thu 7 - Thu 7 Nov-02 SISTA Seminar - Kristiaan Pelckmans
ESAT 00.62
4:00 pm-5:00 pm
LS-SVMlab and Large Scale Modelling''

Kristiaan Pelckmans (KU Leuven, ESAT-SCD-SISTA)

In this presentation an overview is given of black-box modeling techniques related to Least Squares Support Vector Machines. The goal is to give insight in the basic mechanisms of nonparametric kernel based learning, to illustrate these with the newborn Matlab/C toolbox LS-SVMlab and to illustrate on a number of problems. Special attention is given to a new way of handling large datasets. Examples of the applications of these tools in recent projets will be given.

(Talk related to demo track session at NIPS 2002)

Thu 6 - Thu 6 Jun-02 SISTA Seminar - Sven Maerivoet
ESAT - 00.62
4:00 pm-4:30 pm
An introduction to the modeling and simulation of traffic flows

Sven Maerivoet (K.U.Leuven, ESAT-SCD-SISTA)

In this talk we'll give a general introduction to the field of modeling and simulation of real-life traffic flows. We won't discuss traffic control/management but instead we'll focus on some approaches possible for modeling traffic flows. Based on certain considerations, traffic flows can be thought of being macroscopically or microscopically in nature. A small overview of the main variables is given, followed by some of the different models which give rise to modeling traffic flows macroscopically (these include empirical models with their fundamental diagrams and models based on partial differential equations). The main focus, however, will be on the microscopic models, consisting of car-following and lane-changing submodels, which will explicitly model the vehicles' dynamics and interactions. We'll finish by giving two visually interesting examples of possible simulators.

Thu 11 - Thu 11 Apr-02 SISTA Seminar - Lieven De Lathauwer - Jeroen Dehaene
ESAT - 00.62
4:00 pm-6:00 pm
Multilinear algebra and higher-order statistics in signal processing

Lieven De Lathauwer

In this seminar we give an overview of the state-of-the-art of multilinear algebra. We will discuss concept, algorithms and applications for different types of tensor decompositions. We pay some special attention to the problem of Independent Component Analysis and the use of higher-order statistics in signal processing. A related popular subject is simultaneous matrix decompositions.

Local Permutation of products of Bell states and entanglement distillation

Jeroen Dehaene

The seminar will start with a brief introduction to some quantum mechanical notions like qubits, pure and mixed states and entanglement. Then we consider entanglement distillation from a number of copies of a mixed state of a pair of qubits. The distillation protocol is a generalization of a well known two-copy protocol by Bennett et al. This generalization is obtained through the characterization of the group of all locally implementable permutations of the set of products of Bell states as a binary matrix group.

Thu 14 - Thu 14 Feb-02 SISTA Seminar - Katrien De Cock - Christophe Strobbe
ESAT - 00.62
4:00 pm-5:00 pm
Principal angles in system theory, information theory and signal processing

Katrien De Cock (K.U. Leuven, ESAT-SCD-SISTA)

Principal angles between linear subspaces were defined by Camille Jordan in the nineteenth century and statistically interpreted by Hotelling as canonical correlations. In the area of systems and control, the principal angles are used in subspace identification methods and also in model updating and damage location. In the seminar, relations between system theory, information theory and signal processing are established by computing the principal angles between certain subspaces. We start with the definition of the principal angles between linear subspaces and show how the canonical correlations of two stochastic processes can be interpreted as principal angles. Next, we turn to a cepstral distance and norm for autoregressive moving average (ARMA) models. This cepstral norm of a model can be characterized as a function of the principal angles between the row spaces of the controllability matrix of the model and the controllability matrix of the inverse model. We show that the cosines of the angles are equal to the smallest canonical correlations of the output process and its innovation process. Via a geometric property of linear stochastic models, the norm is related to the canonical correlations of the past and the future of the output process and hence to the mutual information of these processes.

The voice-enabled Web: VoiceXML and related standards for telephone access to web applications

Christophe Strobbe (K.U. Leuven, ESAT-SCD-DocArch)

VoiceXML (Voice Extensible Markup Language) is an XML-vocabulary that bridges the public telephone network and the World Wide Web. It is a language for creating audio dialogs that feature synthesized speech, digitized audio, recognition of spoken and DTMF key input, recording of spoken input, telephony and mixed-initiative conversations. VoiceXML browsers provide an interface between a caller on a standard telephone and an application on the web. The advantage of this approach is that companies can build on their existing web infrastructure to add telephone access to their content and services. The VoiceXML 1.0 specification was published in March 2000 by the VoiceXML Forum, and builds upon the work of earlier technologies such as VoxML (Motorola) and SpeechML (IBM). The specification was submitted to the World Wide Web Consortium and became part of a suite of specifications known as the W3C Speech Interface Framework. This framework also comprises the Speech Synthesis Markup Language, the Speech Recognition Grammar Specification, and the Natural Language Semantics Markup Language. All of these languages (including VoiceXML 2.0) are still drafts.

Wed 25 - Wed 25 Apr-01 Collective Behaviour and Complex Systems
ESAT room 00.62
9:00 am-12:00 pm
• Chair: Johan Suykens
09.00-09.50h: Symmetric self-organization in a cellular automaton using a random Gauss-Seidel algorithm - Andre Barbe & Fritz von Haeseler (K.U.Leuven, ESAT/SISTA)
09.50-10.00h: break
10.00-10.50h: Ant colony optimization and swarm intelligence - Marco Dorigo (ULB, IRIDIA)
10.50-11.00h: break
11.00-11.25h: Coupled local minimizers: cooperative search by state synchronization - Johan Suykens (K.U.Leuven, ESAT/SISTA)
11.25-11.50h: Families of scroll grid attractors and synchronization methods - Mustak Yalcin (K.U.Leuven, ESAT/SISTA)
+ n-scrolls music composed by Tom Schouten