Search
mlsplogo MLSP2014
IEEE International Workshop on
Machine Learning for Signal Processing

September 21-24, 2014  Reims, France

ICA and the Big Bang
Jean-Francios Cardoso Jean-François Cardoso
National Center for Scientific Research at the LTCI lab (Laboratoire de Traitement et Communication de l'Information), Paris, France
Personal homepage

Biography:

Jean-François Cardoso (born 1958) is Directeur de Recherche with the French CNRS (Centre National de la Recherche Scientifique) at the LTCI lab (Laboratoire de Traitement et Communication de l'Information, Paris). Since 1989, he has been extensively working on all aspects of blind source separation and independent component analysis. In 2001, he joined the Planck collaboration, a cosmological mission of the European Spatial Agency, for the analysis of Planck data. He developed in particular the blind component separation method which extracted a full-sky high-resolution map of the Cosmic Microwave Background out of the 9 Planck frequency channels. In 2014, he was awarded the CNRS silver medal.

Abstract:

In 2013, the Planck collaboration from the European Spatial Agency released its first results and products, including a full-sky high-resolution map of the Cosmic Microwave Background (CMB) which made the headlines of newspapers worldwide. Indeed, that image is quite literally a snapshot of our Universe in its infancy, when it was a promising but fragile baby, only 380.000 years old. The CMB map is obtained by combining the 9 frequency channels of the Planck satellite in order to separate the precious relic radiation from many other astrophysical emissions. This is a source separation issue which, as it turned out, was best solved by a blind method -- in other words, by Independent Component Analysis-- carefully crafted to deal with the uncertainties in the sky and wild SNR conditions.

The talk will start with an introduction to Big Bang theory, the standard cosmological model, so beautifully supported by Planck results. I will then describe the Planck satellite and sketch its data processing pipeline up to the inference of the cosmological parameters. I will focus on my main contribution to it: blind separation of the CMB from all the other emissions by an ICA algorithm based on a Gaussian spectral likelihood. I will explain why, in my view, the CMB is not sparse in any useful way.


Real-Time Brain Imaging in the Wild
Lars Kai Hansen Lars Kai Hansen
Section for Cognitive Systems, DTU Compute, Technical University of Denmark, Denmark
Personal homepage

Biography:

Lars Kai Hansen received the Ph.D. degree in physics from the University of Copenhagen in 1986. Since 2000 he has held the professorship in digital signal processing at the Technical University of Denmark, where he also heads the Cognitive Systems Section. His research focuses on adaptive signal processing and machine learning with applications in biomedicine, digital media and cognitive systems. He has published more than 280 contributions on these subjects in journals, conferences, and books. The major contributions include introduction and analysis of neural network ensembles (1990-96), adaptation of predictive machine learning methods for neuroimage analysis (1994-), and defining the concept of cognitive components (2005-). His research has been generously funded by the Danish Research Councils, The European Commission, the US National Institutes of Health, and private foundations. He is a present member of the Danish Council for Independent Research - Technology and Production Sciences, and has been active in numerous international research evaluation committees including the Academy of Finland, Vetenskabsraadet, Sweden, INRIA, France, and the US National Institutes of Health. In 2011, he served as Cátedra de Excelencia at Universidad Carlos III de Madrid.

Abstract:

EEG based real-time imaging of human brain function has many potential applications including brain based services and personal precision medicine. In mobile applications real-time imaging is attractive as an element in systems for personal state monitoring and well-being, and in clinical settings for patients who may need imaging under quasi-natural conditions. I will discuss our smartphone based imaging system and current efforts to improve algorithms and interfaces (Stopczynski et al., 2014). Real-time imaging is an interesting challenge at the interface between signal processing and machine learning. Problems related to the ill-posed nature of the EEG imaging escalate in mobile real-time systems and new algorithms and the use of strong priors informed by context and by personal meta-data may be necessary to succeed. I will describe our current research towards fast, robust, yet accurate imaging algorithms. Our work takes a Bayesian approach to sparse activation patterns and includes a Markovian prior for temporally sparse solutions and new inference schemes based on message passing such as the "Variational Garrote" (Kappen, 2011).

Matrix/Tensor based EEG Signal Processing: Algorithms and Applications in Smart Patient Monitoring
Sabine Van Huffel Sabine Van Huffel
KU Leuven, Dept. of Electrical Engineering-ESAT, Stadius Center for Dynamical Systems, Signal Processing and Data Analytics, and iMinds Medical IT Department, Leuven, Belgium
Personal homepage

Biography:

Sabine VAN HUFFEL received the MD in computer science engineering in June 1981, the MD in Biomedical engineering in July 1985 and the Ph.D in electrical engineering in June 1987, all from KU Leuven, Belgium. She is full professor at the department of Electrical Engineering from the Katholieke Universiteit Leuven, Leuven, Belgium. She is (co)supervisor of 40 finished PhDs and 20 ongoing PhDs: mostly all interdisciplinary in collaboration with the University Hospitals Leuven. She has published two monographs and more than 350 articles in refereed international journals. In April 2013 she received an honorary doctorate from Eindhoven University of Technology, together with an appointment as a Distinguished professor from January 1, 2014 to January 1, 2018. She is heading the Biomedical Data Processing Research Group (BIOMED) within the Stadius Center for Dynamical Systems, Signal Processing and Data Analytics. BIOMED focuses on the development of numerically reliable and robust algorithms for improving medical diagnostics. Within BIOMED both fundamental/theoretical and application-oriented research is performed in domains such as classification and prediction modeling, and (non)linear signal analysis and pattern recognition. Applications under study include cancer diagnosis (ovarian, breast, endometrial, brain), early pregnancy diagnosis, in-vivo Magnetic Resonance Spectroscopic (MRS) data quantification and imaging, heart-rate variability analysis, and integrated EEG-functional MRI (fMRI) data processing. BIOMED is also part of the iMinds Future Health Department, the aim of which is to do research to support decision support for clinicians, patients, and policy makers. For more information, see http://www.esat.kuleuven.be/stadius/.

Abstract:

In EEG signal processing, the aim is to extract clinically relevant information (e.g rhythms, evoked potentials, bursts, seizure activity patterns) from scalp recordings in order to enable an improved medical diagnosis. Typically, EEG data are affected by artefacts (eye, muscle, etc.) and are of low quality, largely due to the non-invasive and nonobtrusive nature of the measurement process. Accurate and automated quantification of this information requires an ingenious combination of adequate pretreatment of the data (e.g. artefact removal), feature selection, pattern recognition, decision support, up to the embedding of these advanced tools into user-friendly user interfaces to be used by clinicians.

The underlying computational signal processing problems can be solved by making use of matrix and tensor decompositions. In particular, it is shown how Principal Component Analysis, Canonical Correlation Analysis, Independent component Analysis, Parallel Factor Analysis and Block Component Analysis, can be used as building blocks of higher-level signal processing algorithms. In addition, the application of these decompositions and their benefits will be illustrated in a variety of case studies, including epileptic seizure onset localisation using adult and neonatal scalp EEG and Event-related potential analysis during simultaneous EEG-fMRI acquisition

Powered by CONWIZ, © Copyright 2014. Page maintained by Jan Larsen (Email)