History
of the laboratory
I developed my first research activities in the image
processing and pattern recognition laboratory (TIRF),
which became, in January 1998, a joint unit of CNRS, the
images and signals laboratory (LIS), by merging two
laboratories, TIRF and CEPHAG. In January 2007, by
merging of three laboratories: LIS, LAG (Automatic control
lab.) and ICP (Speech lab.), is born the large laboratory
GIPSA-lab.
I first worked under the supervision of Prof. Jeanny
Hérault on modelisation of neural networks and design
of artificial neural networks algorithms,
with applications in signal and ilmage processing.
Since end of 80's, I develop my own research activities,
with my PhD students and collaborators, mainly on source
separation and independent component analysis (ICA),
sparse representation and their applications.
After theoretical contributions on linear
instantaneous mixtures (from 1984), convolutive mixture
(from 1987, but especially after 1990), I was interested
by source separation in nonlinear mixtures (especially
from 1996). In addition, under the impulse of my colleague
and friend, Prof. Massoud Babaie-Zadeh from Sharif
University of Technology (Tehran, Iran), I also
addressed sparse representation and sparse sourse
separation in under-determined mixtures since 2008.
Current
researches
Since 1981, my research
activities are focused on statistical signal processing
issues strongly related with machine learning. Very early,
I investigated artificial neural networks, how they
process information. For understanding how vertebrate
brain are able to code/decode their own motions, with
Prof. J. Hérault and Dr. B. Ans, I developped the concept
of source separation and contributed in its theoretical
fundations. I also applied the results in many
applications, especially in audio-visual source
separation, in biomedical engineering (ECG, EEG, etc.),
hyperspectral imaging and chemical sensing. More recently,
I investigated sparse representations, dictionary
learning, and data fusion from a theoretical perspective
in the framework of Big Data, which are typically
multimodal, heterogeneous, high-dimension, with again
applications in brain imaging, remote sensing, chemical
sensor array, etc. With Dr. M. Congedo, I also work on
brain-computer interfaces, and especially in
representation and processing on Riemanian manifolds.
During the last past 6 years, I mainly worked with the
support of
European
Research Council, with an ERC Advanced Grant for the
project
Challenges
on Extraction and Separation of sources (CHESS). The
three main challenges are:
- Source separation for multimodal recordings;
- Source separation for nonlinear mixtures;
- Source separation with data in very large or very
small dimensions.
A summary of the results obtained during this project can
be found in the
final
CHESS project report.
The main applications I currently address, and which are
necessary both for validating theoretical results and for
raising new relevant issues with realistic assumptions,
are the following:
- Smart chemical and gas sensor arrays, based on source
separation. Preliminary contributions have been done in
cooperation with LAAS (Toulouse, Dr. P. Temple-Boyer)
and Universitat Politecnica de Catalunya
(Barcelona, Prof. J. Cabestany, Dr. Bermejo) in 2000's.
Currently, I work mainly with Dr. L. Duarte (Univ. of
Campinas, Brazil) on theoretical issues, especially due
to nonlinearity of sensors, and with Dr. P. Grangeat
(CEA, Grenoble) on gas sensors. This application domain
is especially very interesting for addressing nonlinear
models of mixtures.
- ICA and advanced signal processing methods, applied to
biomedical signals signals with Dr. B. Rivet and Dr. R.
Phlypo (GIPSA-lab). I especially consider non invasive
fetus ECG extraction with Prof. M. Shamsollahi (Sharif
Univ. of Technology), Prof. Reza Sameni (Shiraz Univ.),
processing of EEG signals and MRI images with Prof.
Soltanian-Zadeh (Univ. of Tehran) in cooperation with
Prof. Ph. Kahane and Dr. A. Depaulis in Grenoble
Institut of Neurosciences, brain-computer interfaces
with Dr. Marco Congedo (GIPSA-lab).
- Unmixing of hyperspectral images with Prof. Chanussot
(GIPSA-lab) et al. After analysis of recordings of the
planet Mars, we addressed different problems in
hyperspectral unmixing, especially considering
structured sparsity and spectral variability.
Team
and Cooperations
During the last past 6
years, these works have been developed through strong
collaborations:
- with permanent researchers in GIPSA-lab: Dr.
Bertrand Rivet, Dr. Ronald Phlypo, Dr. Marco
Congedo, Prof. Jocelyn Chanussot, Prof. Olivier
Michel and Dr. Pierre-Olivier Amblard,
- with permanent researchers abroad: Prof. Tulay
Adali (Univ. of Baltimore County, USA), Prof. M.
Babaie-Zadeh, Prof. Shamsollahi (Sharif University
of Technology,) and Prof. Leonardo Duarte
(University of Campinas),
- with post-doc researchers: Dr. Dana Lahat, Dr.
Simon Henrot, Dr. Paolo Zanini, Dr. Anthony
Bourrier,
- and with many PhD students: Farnaz Sedighin,
Mostada Sadeghi, Fateme Ghayen, Fardin Afdideh,
Saloua Chlaily, Bahram Esandoust, Saeed Akhavan,
Pedro Rodrigues.
Grenoble Images Parole
Signal Automatique
laboratoire
UMR 5216 CNRS - Grenoble INP -
Université Joseph Fourier - Université Stendhal