ERNSI WS System Identification 2008


The seventeenth ERNSI Workshop in System Identification will be held October 1-3 2008 in Sigtuna at the Sigtuna Foundation,


Sigtuna is very close to the Stockholm Arlanda Airport (20 minutes by taxi), and is a small old cultural city. The conference place is also very special and in line with ERNSI traditions.


The workshop is sponsored by the VR Linnaeus Research Center ACCESS (Autonomic Complex Communication nEtworks Signals and Systems) at KTH,, the Center for Industrial and Applied Mathematics, CIAM at KTH and Uppsala University.




Final Program


Wednesday October 1


11:00 - 12:00



12:00 - 13:00



13:00 - 13:15



13:15 - 14:00

Fredrik Gustafsson invited speaker (ULIN)

Modeling for Kalman filtering


14:00 - 14.30

Johan Schoukens, Kurt Barbé, Laurent Vanbeylen and Rik Pintelon (VUB)

Study of the nonlinear induced variability in linear system identification

14:30 - 15:00

Marco Campi (UNIBS)

Interval prediction models: Identification and reliability

15:00 - 16:30

Coffee & Poster session






Poster 1

Cristian Rojas (KTH)

Fundamental limitations in system identification and their application to experiment design


Poster 2

Kurt Barbé (VUB)

Using ANOVA in a microwave Round-robin comparison


Poster 3

Agnes Rensfelt and Torsten Söderström (UPPSALA)

Validation of wave propagation models used in nonparametric identification of viscoelastic


Poster 4

John Lataire and Rik Pintelon (VUB)

How fast is a time-varying system varying?


Poster 5

Rogier Blom, Paul Van den Hof, Hans Langen and Rob Munnig Schmidt (TUD)

Multivariable frequency response function estimation of a micro-milling spindle with active magnetic bearings


Poster 6

Eric Wernholt (ULIN)

Frequency-domain identification of industrial robots


Poster 7

Jan-Willem van Wingerden and Michel Verhaegen (TUD)


The analogy between LPV and LTI predictor-based subspace identification


Poster 8

Enrico Avventi and Per Enqvist (KTH)

Approximative linear and logarithmic interpolation of spectra

16:30 - 18:30

White paper session chaired by Håkan Hjalmarsson (KTH)

A road map for system identification






Thursday October 2

08:15 - 08:45

Ben Hanzon (UCC)

Subdiagonal pivot structures and associated canonical forms under state isometries

08:45 - 09:15

Wolfgang Scherrer (TUW)

Rational approximation of fractionally integrated processes

09:15 - 09:45

Tzvetan Ivanov (UCL)

Applications of real rational modules in system identification

09:45 - 11:15

Coffee & Poster session






Poster 1

Cristian Rojas, Märta Barenthin Syberg, James Welsh and Håkan Hjalmarsson (KTH)

The cost of complexity in system identification


Poster 2

Philippe Dreesen (KULEUVEN)

The Riemannian singular value decomposition in dynamic system identification


Poster 3

Laurent Vanbeylen, Rik Pintelon, Pieter de Groen (VUB)

Blind maximum likelihood identification of Wiener systems with measurement noise


Poster 4

Torsten Söderström, Magnus Mossberg and Mei Hong (UPPSALA)

A covariance matching approach for identifying errors-in-variables systems


Poster 5

Ludwig De Locht, Gerd Vandersteen and Yves Rolain

Identification of the nonlinear circuit contributions caused by multitone excitation


Poster 6

Märta Barenthin Syberg, Bo Wahlberg, Håkan Hjalmarsson and Mathias Barkhagen (KTH)

L2 gain estimation in noise


Poster 7

Bo Wahlberg (KTH)

On identification of cascade systems


Poster 8

Roland Toth (TUD)

A prediction-error framework for LPV systems




11:15 - 12:15

Timo Koski, invited speaker (KTH)

On learning of structures for Bayesian networks

12:15 - 13.30



13:30 - 14:00

Vincent Laurin, Marion Gilson, Hugues Garnier and Peter Young (UHP-NANCY)


An IV-based method for non-linear continuous-time Hammerstein model identification: Application to rainfall-flow modelling


14:00 - 14:30

László Gerencsér and Vilmos Prokaj (SZTAKI)

Stability of hybrid linear stochastic



14:30 - 15:00

Tillmann Falck  (KULEUVEN)

Robust kernel based regression in SOCP and LS formulations for perturbation analysis


Bus to Uppsala


16:00 - 17:00

Visit Gustavianum


17:00 -18:00

Free time in Uppsala



Bus from Uppsala







Friday October 3


08:30 - 09:00

Christian Lyzell, Martin Enqvist and Lennart Ljung (ULIN)

An algebraic approach to convexification of system identification problems


09:00 - 09:30

Huseyin Akcay (UCL)

Synthesis of complete orthonormal fractional bases


09:30 - 10:00

Gianluigi Pillonetto, Alessandro Chiuso and Giuseppe De Nicolao (UNIPD)


Predictor estimation via Gaussian regression: theory and applications

10:00 - 10:30




10:30 - 11:30

 Peter Grunwald, invited speaker (CWI)

The catch-up phenomenon in model selection and prediction


11:30 - 12:00

Thomas Schön (ULIN)

A new algorithm for calibrating a combined camera and IMU sensor unit

12:00 - 13:15




13:15 - 13.45

Martin Enqvist (ULIN)

An improved weighting method for initialization of Hammerstein or Wiener system identification algorithms


13:45 - 14:15

Jonas Sjöberg (VUB/CHALMERS), Per-Olof Gutman, Makul Agerwal, Mike Bax


Tuning of a PID-controller for the Furuta pendulum using a sequence of identifications of linearized time-varying models


14:15 - 14:45

Henrik Ohlsson (ULIN)

Manifold-constrained regressors in system identification


14:45 - 15:00





Invited Presentations


Fredrik Gustafsson, ULIN, Sweden

Modelling for Kalman filtering

Abstract: Modeling for filtering is a rather unexplored area in contrast to identification for control. Even if the true system is known, it is not clear which linear model is best suited to use in a Kalman filter. In contrast to control applications, the predictive ability of the model is more important than the dynamics from the input. The presentation aims at maximizing confusion by providing no solutions but only a number of practical and toy examples, where the best model order is larger or smaller than the true system, and where the input dynamics is not used.



Timo Koski, KTH, Sweden

On learning of structures for Bayesian networks


Peter Grundwald, CWI, the Netherlands

The Catch-Up Phenomenon in Model Selection and Prediction


Abstract: Standard Bayesian model selection/averaging sometimes learn too slowly: there exist other learning methods that lead to better predictions based on less data. We give a novel analysis of this "catch-up" phenomenon.

We resolve a long-standing debate in statistics, known as the AIC-BIC dilemma: model selection/averaging methods like BIC, Bayes, and MDL are consistent (they eventually infer the correct model) but, when used for prediction, the rate at which predictions improve can be suboptimal. Methods like AIC and leave-one-out cross-validation are inconsistent but typically converge at the optimal rate. Both AIC and BIC may be viewed as striking a trade-off between a model's goodness-of-fit and its complexity, but under AIC, the 'complexity' of any given model is typically smaller than under BIC.

We give a novel analysis of the slow convergence of the BIC/Bayesian-type methods. Based on this analysis, we propose the switching method, a modification of Bayesian model averaging that achieves both consistency and minimax optimal convergence rates. The method is related to expert-tracking algorithms developed in the COLT literature, and has time complexity comparable to Bayes. Experiments with nonparametric density estimation confirm that our large-sample theoretical results also hold in practice in small samples.

Joint work with T. van Erven and S. de Rooij.