CHESS

Monday, a special Chess meeting, with only a talk, for the beginning of the year and the leaving of Alexandre Akzenov.

It will take place in room Chartreuse at 4pm with some galette des Rois.

For organization purpose, please fill the Doodle : http://doodle.com/poll/e2e4gzzrh85a5ems

The next CHESS meeting will be tomorrow, Monday January 23, from 10:00 to 11:30 (French time). It will be in room Chartreuse, and with Skype connection with the team of Prof. Babaie-Zadeh in Tehran.

There will be two talks, the first given by Fateme GHAYEM and the second by Saloua CHLAILY.
Fateme GHAYEM
Title: Sparse Signal Recovery via Iterative Sparsification-Projection: A Closer Look and Accelerated Extensions

Abstract: This paper studies a recently proposed family of algorithms, called iterative sparsification-projection (ISP), for
recovery of sparse signals. The ISP algorithms are motivated by the idea of the smoothed `0 (SL0) method, which approximates the L0 norm with a continuous and differentiable function and solves an error-constrained problem by utilizing the well-known graduated non-convexity scheme. The ISP algorithms generalize this idea to non-smooth sparsity inducing functions. Although these algorithms have shown promising performance, some important aspects of them have not yet been fully investigated. In particular, there is no direct derivation and convergence analysis for the non-smooth case. In this paper, we are going to demonstrate the potential advantages of the ISP algorithms by revealing some interesting properties of them along with proposing acceleration schemes to further enhance their recovery performance. More precisely, it is shown that the SL0 shrinkage is in fact a smooth interpolation between hard and soft thresholdings.
Furthermore, the close connection of the SL0 shrinkage with a well-known shrinkage function, called smoothly clipped absolute deviation (SCAD), is discussed. Our simulations indicate that using the SCAD shrinkage in ISP leads to a significant improvement of the performance relative to the use of other shrinkage functions. As another contribution, using an alternating minimization penalty method, we directly derive the ISP algorithms for non-smooth sparsity-promoting functions, including SCAD and hard thresholding. Moreover, we propose accelerated extensions of the ISP algorithms for
both smooth and non-smooth cases, and establish convergence to critical points for the resulting algorithms. Our extensive simulations verify that the new accelerated algorithms considerably outperform their plain versions and some wellknown and recently proposed algorithms.

Saloua CHLAILY
Title: Information-Estimation relationship in Mismatched Gaussian channels
Abstract: In this paper, we investigated the connection between information and estimation measures in mismatch modeling contexts. Additionally to the input prior mismatch, the novelty of this paper is to take into account the noise mismatch which has not been studied yet and deserves to be explored for some applications. A new relation between relative entropy and excess mean square error is stated. Finally, an example illustrates the impact of model mismatches on estimation accuracy.

The next CHESS meeting will take place on Monday, January 9, at 10:00 am in room Chartreuse.

For this first meeting of the year, Christian suggest another kind of presentations, in order to everybody know what are doing the others.
So, he would like that every PhD and post-doc students involved in the CHESS project present briefly (max 2-3 slides, and about 5 minutes) what she/he is interested in and investigating.

We also try to have a Skype connexions with Sharif University of Technology, so that PhD and MSc students from Sharif can also present their current and future project.
Unfortunately, for our Brazilian colleagues from Campinas, it will be probably too early for joining us.

We apologize for this late announcement, but we believe that it does not take you long time for preparing a few slides.

The next CHESS meeting will take place on Monday, December 12, at 10:00 am in room Chartreuse.

There will be 3 talks:

Victor MAURANDI
Title: Multimodal Approach to Remove Ocular Artifacts from EEG Signals Using Multiple Measurement Vectors
Abstract: This presentation deals with the extraction of eye-movement artifacts from EEG data using a multimodal approach. The gaze signals, recorded by an eye-tracker, share a similar temporal structure with the artifacts induced in EEG recordings by ocular movements. The proposed approach consists in estimating this specific common structure using Multiple Measurement Vectors which is then used to denoise the EEG data. This method can be used on single trial data and can be extended to multitrial data subject to some additional preprocessing. Finally, the proposed method is applied to gaze and EEG experimental data and is compared with some popular algorithms for eye movement artifact correction from the literature.

Florent BOUCHARD
Title: Approximate Joint Diagonalization according to the Natural Riemannian Distance"
Abstract: Here, we propose for the first time an approximate joint diagonalization (AJD) method based on the natural Riemannian distance of Hermitian positive definite matrices. We turn the AJD problem into an optimization problem with a Riemannian criterion and we developp a framework to optimize it. The originality of this criterion arises from the diagonal form it targets. We compare the performance of our Riemannian criterion to the classical ones based on the Frobenius norm and the log-det divergence, on both simulated data and real electroencephalographic (EEG) signals. Simulated data show that the Riemannian criterion is more accurate and allows faster convergence in terms of iterations. It also performs well on real data, suggesting that this new approach may be useful in other practical applications.

Pierre PIGNEDE
Title: Interface for codes and data public access.

The last CHESS meeting of 2016 will be followed at 11:30 by a friendly drink.

Thursday, the 21th july in room Chartreuse at 3pm.

Title: "Smoothed L0 (SL0): A fast algorithm for finding the sparse solution of an underdetermined system of linear equations"

Abstract: "In this talk, the problem of finding the sparse solution of an underdetermined systems of linear equations (the "sparse recovery" problem) is considered. We first discuss the significance of this problem by mentioning a few of its applications (including Compressed Sensing), and we point out the uniqueness of such a solution. Then, after pointing out that directly finding such a solution is computationally impossible, we briefly review the main different ideas for finding such a solution, including Matching Pursuit idea and Basis Pursuit idea (minimizing the L1 norm). Finally, we present an approach, called Smoothed L0 (SL0), which tries to directly minimize the L0 norm. We will see that this idea will result in an algorithm which is 2 to 3 orders of magnitude faster than L1-magic (a well-known and fast implementation for Basis Pursuit

CHESS and DECODA organize there hackaton beginning of july :

Summary of the two days

• On the first day during the 1st and 2nd Hackathon sessions, people will group themselves around a topic of choice. The goal is to discuss/program/draw/invent/proof around a central problem in an environment that stimulates cooperation, collaberation, joint thinking, paving the road to a (partial) solution.

• On day 2 the groups can be reorganised, keeping in mind that a comfortable group size is about 4 persons. The goal is to crossfeed among projects for some, whilst keeping the possibility to work during the full two days on the same single project for others. Konstantin and Ronald will organise a poll during the Hackathon Summary Day 1 and reorganise the groups accordingly.

Both days will welcome tutorials (big thanks to Laurent Condat and Simon Barthelmé) with a focus on optimisation for signal/image processing.

more informations here

The next CHESS meeting will held today, Tuesday March 15, in room Mont-Blanc, at 10:00 AM.

1) There will be first two talks

Lucas Drumetz (Univ. Grenoble Alpes)
Title: Optimization tools for signal and image processing: application to spectral unmixing of hyperspectral images (2/2)

This second part (the first part on optimization methods was presented on March 15) of the talk will be focused on applications in spectral unmixing of hyperspectral imaging.

Paolo Zanini (Univ. Grenoble Alpes)
Title: Parameters estimate of Riemannian Gaussian distribution in the manifold of covariance matrices

The study of P(m) , the manifold of m × m Symmetric Positive Definite matrices, has recently become widely popular in many engineering applications, like radar signal processing, mechanics, computer vision, image processing, and medical imaging. A large body of literature is devoted to the barycentre of a set of points in P(m) and the concept of barycentre has become essential to many applications and procedures, for instance classification of SPD matrices. However this concept is often used alone in order to define and characterize a set of points. Less attention is paid to the characterization of the shape of samples in the manifold, or to the definition of a probabilistic model, to represent the statistical variability of data in Pm. Here we consider Gaussian distributions and mixtures of Gaussian distributions on P(m). In particular we deal with parameter estimation of such distributions. This problem, while it is simple in the manifold P(2), becomes harder for higher dimensions, since there are some quantities involved whose analytic expression is difficult to derive. In this paper we introduce a smooth estimate of these quantities using convex cubic spline, and we show that in this case the parameters estimate is coherent with theoretical results. We also present some simulations and a real EEG data analysis

2) Information/exhibition on Redmine (P. Pignède and R. Phlypo)

3) Future event: scientific DECODA/CHESS days in July 2016 (K. Usevitch and R. Phlypo)

4) "Help" talk, if any ???

The next CHESS meeting will held today, Tuesday March 15, in room Mont-Blanc, at 10:00 AM.

There will be two talks, the first one presented by Bahram Ehsandoust, and the second one by Lucas Drumetz.

Bahram Ehsandoust (Sharif Univ. of Technology and Univ. Grenoble Alpes)
Title: Nonlinear Blind Source Separation for Sparse Sources

Lucas Drumetz (Univ. Grenoble Alpes)
Title: Optimization tools for signal and image processing: application to spectral unmixing of hyperspectral images

There will be two talks, the first one presented by Denis Fantinato, and the second one by Raphaëlle Roy.

Raphaëlle Roy
Title The DynEmo project: Eye-fixation related potentials of emotional facial expressions

Abstract The processing of emotional facial expressions (EFE) elicits specific evoked brain responses that enable us to better understand at what stage the differentiation of the valence and nature of these emotions occurs. Typically, these evoked responses are investigated using the event-related potential technique (ERP). However, in order to assess more subtly the time course of EFE processing, it seems more appropriate to use eye-fixation related potentials (EFRP) thanks to the co-registration of electro-encephalography and oculometry. Moreover, only this technique can allow an investigation of the impact of specific face features, such as the eyes, on the brain responses to EFE. Objective: This study was designed to investigate how the evoked potentials elicited by natural emotional facial expressions are impacted by four emotion conditions (for both ERPs and EFRPs) –happiness, surprise, disgust, neutral- as well as the face region of interest – eyes, nose and mouth – in the case of EFRPs. Methods: Twenty-four participants underwent the experiment. They had a first session during which they passively explored the pictures of natural but standardized EFE of the DynEmo database, and a second one during which they had to label them. Main results: The eyes elicited significantly more fixations but lower amplitudes for all components than the nose, which in turn elicited more fixations and lower components’ amplitude than the mouth. Moreover, the disgust condition elicited more fixations on the nose and less on the eyes than the other emotions, and the happiness and surprise conditions elicited more fixations on the mouth than the other emotions. At the electrophysiological level, the distinction between emotions appeared around 220 ms post-stimulation for the ERPs with significantly different amplitudes for the P2P3 complex and the LPP component, and even earlier for the EFRPs with modulations of the N170 component, mostly at the left frontal electrode sites. The face region was critical as the mouth elicited higher components’ amplitude than the nose, and the nose than the eyes at central and parieto-occipital sites. There was also an interaction between the face region and the emotion, with notably a distinction between surprise and disgust on the mouth for the P2P3 complex.

Denis Fantinato
Title: From Blind Deconvolution to Nonlinear Blind Source Separation in the Context of Statistically Dependent Sources

Abstract: This presentation will discuss about two problems in the signal processing area: that of blind deconvolution and that of nonlinear blind source separation, both considering statistically (temporally) dependent sources.
In the former case, the problem admits solution – based on results obtained in my PhD research – if certain a priori dependence information is at disposal, which is a valid assumption in some applications, like that of channel equalization. This idea will be presented under the light of the matching of multivariate distributions, using the Parzen window method for PDF estimation.
In the second case, for my stage at GIPSA, we focus on a new perspective of the field: the use of the derivatives of the mixtures. Interestingly, a possible approach is to explore the statistical independence among the sources by jointly considering the mixtures and its derivatives – an idea that could be accomplished through the use of multivariate distributions. Some other possibilities of work will be presented for discussion/suggestion.

Talk (~40mn) :

Dana Lahat++ : When blind source separation meets data fusion

Abstract: Blind source separation (BSS) is a well-established framework for data analysis. Independent component analysis (ICA) and algorithms such as JADE, SOBI, FastICA and InfoMax, designed in the 1990's, are now regarded as standard analytical tools in a broad range of domains, and are almost synonymous with BSS. However, for various types of real-life data, their underlying assumptions are too restrictive or insufficient. In fact, from the very early days of BSS, other types of models have been proposed that can better accommodate the richer properties of the data. In this talk, we focus on two of these directions: relaxing the assumption of independence within a single data set, and allowing statistical dependence between different data sets. The latter opens new horizons and opportunities in data fusion and multiset data analysis. We present a few recently-proposed models, algorithms, and new results on uniqueness.