Pierre Vandergheynst

photo placeholder image

Vice-President for Education

pierre.vandergheynst@epfl.ch +41 21 693 56 45 http://lts2.epfl.ch

EPFL VPE
CE 3 304 (Centre Est)
Station 1
CH-1015 Lausanne

EPFL VPE
CE 3 304 (Centre Est)
Station 1
CH-1015 Lausanne

EPFL STI IEL LTS2
ELE 235 (Bâtiment ELE)
Station 11
CH-1015 Lausanne

EPFL IC-DO
ELB 114 (Bâtiment ELB)
Station 11
CH-1015 Lausanne

Unit: SEL-ENS

Unit: FCUE

Web site: http://cds.epfl.ch/
Unit: CDS

Unit: CDOCT

perm_contact_calendarvCard
Administrative data

Fields of expertise

data science
machine learning
computational harmonic analysis
inverse problems
compressive sensing
computer vision

Publications

Infoscience

Other publications

P. Frossard, P. Vandergheynst, R. Figueras i Ventura and M. Kunt
IEEE Transactions on Signal Processing, Vol. 52, No 2, February 2004
A Posteriori Quantization of Progressive Matching Pursuit Streams
Hagmann P , Thiran J , Jonasson L , Vandergheynst P , Clarke S , Maeder P and Meuli R
, Neuroimage, Vol. 19, No 3, pp. 545-554, July 2003, 2003
DTI mapping of human brain connectivity: statistical fibre tracking and virtual dissection
N. Aspert, T. Ebrahimi and P. Vandergheynst
Computer Aided Geometric Design, Vol. 20, No 3, 2003
Non-linear subdivision using local spherical coordinates
Antoine J , Demanet L , Jacques L and Vandergheynst P
Appl. Comp. Harmonic Analysis, Vol. 13, No 3, pp. 177-200, November 2002, 2002
Wavelets on the sphere : Implementation and approximations
Vandergheynst P and Frossard P
Signal Processing, Vol. 82, No 11, pp. 1517-1518, November (2002), 2002
Special issue on image and video coding beyond standards

Research

Topics

Data nowadays come in overwhelming volume. In order to cope with this deluge, we explore and use the benefits of geometry and symmetry in higher dimensional data. But volume is not the only problem: data models are also increasingly complex, mixing various components. We thus use sparse representations and dictionaries as dimensionality reduction tools to dig out information from complicated high-dimensional datasets and multichannel signals, or to model complex behaviours in more classical signals. Finally data can also be complex because they are collected on surfaces, or more generally manifolds, or because they are not scalar-valued. We thus explore extensions of Computational Harmonic Analysis in higher dimensions, in complex geometries, on graphs, networks or for non-scalar data.