Prof. Dr. Ann Dooms

Postdoctoral Researcher/Part-time ZAP

Welcome to the home page of Ann Dooms,
member of IBBT/ETRO-IRIS (WAM) and CAMP at the Department of Electronics and Informatics (ETRO),
Faculty of Engineering of the Vrije Universiteit Brussel, Belgium.

Former member of ALGB at the Department of Mathematics (DWIS),
Faculty of Sciences of the Vrije Universiteit Brussel, Belgium.

My view


Pleinlaan 2
1050 Brussel

Office:  PL9 2.29

Telephone: +32(0)2-629 16 69

Fax: +32(0)2-629 17 00




This semester, a weekly CAMP seminar on Mondays from 16-18 will be organized with topics including inverse problems and regularization (e.g. in medical imaging and seismic tomography), scalable image coding, watermarking, error correcting codes, ...

Monday 22nd February 2010, 16-18, 6G324

Title: Catching features with a lasso or an elastic net?

by Prof. Dr. Christine De Mol (Université Libre de Bruxelles, Dept Math and ECARES Vrije Universiteit Brussel, CAMP)

I will describe feature selection methods based on penalized regression, and in particular the "lasso" (L1 penalty) and the "elastic net" (combined L1 and L2 penalty). I will summarize the theoretical background in the framework of learning theory and focus on two specific applications: face detection in computer vision and gene selection in bioinformatics.

Christine De Mol holds a Ph.D. in Physics (1979) and a habilitation degree in Mathematical Physics (1992) from ULB. Since 1975, she has held several research positions with the Belgian National Fund for Scientific Research (FNRS) that she left in 1998 as a Honorary Research Director to become a full-time Professor at ULB. She has held several visiting positions (Universities of London, Rome, Montpellier, Paris-Sud, Genoa).Her research interests in applied mathematics include inverse problems, sparsity-enforcing regularization theory, wavelets and applications, learning theory, analysis and forecasting of time series, portfolio theory.


The first seminar in this series (Monday 15th February 2010, Pleinlaan 9 Marconi) was given by

Johann Briffa
from University of Surrey (Guildford, UK) on 'Insertion/Deletion' correcting codes.

The Insertion/Deletion channel is a well-known topic in coding theory, but
very little research has actually been carried out to study practical codes
for large block lengths.
Recently, such codes have found novel applications in digital watermarking,
and a need for further research in coding theory is becoming evident.

In this presentation, we introduce the problem of robust watermarking,
the challenge of recovering from local geometric attacks, and the potential
application of insertion/deletion correcting codes in this context.
This is followed by an overview of existing codes for insertion/deletion
correction, focusing on the scheme due to Davey and MacKay.

The original Davey-MacKay construction is discussed, and our contributions
to the study of these codes is detailed.
Most importantly, we show how the decoding algorithm can be improved, to
obtain better decoding performance with the same encoder.
We also discuss the design of the inner code in the Davey-MacKay system,
and show that some care should be taken in its design.
Our improved decoding algorithm does not require the codewords to be
low-weight, and we show that codes with larger minimum distance give further
improvements with the new decoder on medium and good channels, i.e. with
insertion and deletion probabilities lower than about 10%.

Everybody is cordially invited!