Tutoriel 1: Vue d'ensemble sur l'analyse de données pour l'astronomie gravitationnelle : principes, algorithmes et données ouvertes

• Florent Robinet (LAL - Orsay)

Room: Univ Paris Diderot -- UFR Physique, Amphi Pierre-Gilles de Gennes

Tutoriel 2: Introduction à l'apprentissage statistique

• Stéphane Canu (LITIS, Rouen)

Room: Univ Paris Diderot -- UFR Physique, Amphi Pierre-Gilles de Gennes

Overview about machine learning in gravitational wave astronomy

• Agata Trovato (INFN - LNS)

Room: Univ Paris Diderot -- UFR Physique, Amphi Pierre-Gilles de Gennes We will give an overview of the current projects using machine learning in the context of gravitational wave astronomy, the results obtained so far and their potential impact.

Tackling data analysis challenges in astrophysics with sparse matrix factorization methods

• Jerôme Bobin (CEA Irfu)

Room: Univ Paris Diderot -- UFR Physique, Amphi Pierre-Gilles de Gennes In this presentation, we review some important applications of sparse matrix factorization (SMF) methods in astrophysical. In this context, such methods are playing a key role in the analysis of multi-frequency observations to disentangle between underlying astrophysical components. The first application will deal with the estimation of the most ancient relic radiation from the Big Bang, a.k.a. Cosmological Microwave Background (CMB), from the Planck data. In this scope, we will emphasize on how sparse matrix factorization is performed on microwave full-sky data and how sparsity largely helps disentangling between the sought-after CMB and the spurious "foreground" components. The second example will emphasize on the extension of these methods to analyse radio-interferometric measurements, in preparation of the forthcoming very large radio-telescopes such as the Square Kilometer Array (SKA). In contrast to the previous case, component separation has to be performed from incomplete data: the data (or so called visibilities) are equivalent to a subsampling in the Fourier space. In plain english, this means that one has access to only few samples. In this case, we will introduce an extension of SMF that is specifically tailored to deal with incomplete data, with illustrations from radio-interferometric measurements.

Tutoriel 3: Revue et état de l'art sur les méthodes d'estimation bayésienne

• Nicolas Dobigeon (IRIT/INP-ENSEEIHT & IUF, Toulouse)

Room: Univ Paris Diderot -- UFR Physique, Amphi Pierre-Gilles de Gennes

A Hamiltonian Monte Carlo sampler for compact binary sources

• Marc Arène (CNRS/IN2P3 APC)

Room: Univ Paris Diderot -- UFR Physique, Amphi Pierre-Gilles de Gennes A major activity of the LIGO-Virgo collaboration is to build algorithms able to infer from the detected gravitational wave signals the posterior distributions of the parameters defining their sources: angles in the sky, distance from us, masses etc. Current algorithms like MCMC and Nested Sampling have already demonstrated with success their ability to do so during the first and second run of observations of the detectors. However with the third round of observations coming in 2019 we expect much higher rates of detections and hence the need for faster algorithms. I will present here how a Hamiltonian Monte Carlo sampler can respond to this challenge when designed appropriately.

Non-parametric Characterization of Gravitational-Wave Polarizations

• Julien Flamant (Cristal Lille)

Room: Univ Paris Diderot -- UFR Physique, Amphi Pierre-Gilles de Gennes Gravitational waves are polarized. Their polarization is essential to characterize the physical and dynamical properties of the source i.e., a coalescing binary of two compact objects such as black holes or neutron stars. Observations with two or more non coaligned detectors like Virgo and LIGO allow to reconstruct the two polarization components usually denoted by h+(t) and h(t). The amplitude and phase relationship between the two components is related to the source orientation with respect to the observer. Therefore the evolution of the polarization pattern provides evidence for changes in the orientation due to precession or nutation of the binary. Usually, some specific direct dynamical model is exploited to identify the physical parameters of such binaries. Recently, a new framework for the time-frequency analysis of bivariate signals based on a quaternion Fourier transform has been introduced in [1]. It permits to analyze the bivariate signal combining h+(t) and h(t) by defining its quaternion embedding as well as a set of non-parametric observables, namely Stokes parameters. These parameters are remarkably capable of measuring fine properties of the source, in particular by deciphering precession, without close bounds to a specific dynamical model.

Data analysis challenges with LISA: current status and prospects

• Nikolaos Karnesis (CNRS/IN2P3 APC)

Room: Univ Paris Diderot -- UFR Physique, Amphi Pierre-Gilles de Gennes This talk will review the key issues that has to be addressed to extract the full science potential from the data of the LISA space mission. We will describe the LISA data challenges which will provide the main framework to develop and test data analysis algorithms.

Analyse temps fréquence des ondes gravitationnelles en utilisant synchrosqueezing d’ordre supérieur

• Duong-Hung Pham (Gipsa-lab)

Room: Univ Paris Diderot -- UFR Physique, Amphi Pierre-Gilles de Gennes Cet exposé présentera une analyse temps fréquence des ondes gravitationnelles en utilisant une nouvelle méthode: le synchrosqueezing d’ordre supérieur fondé sur la transformée de Fourier à court terme (TFCT). Tout d’abord, nous rappellerons le modèle mathématique de signal multicomposantes (SMCs) defini comme une superposition de modes modulés à la fois en amplitude et en fréquence (appelés modes AM-FMs). Pour l’analyse de tels signaux, la TFCT est une méthode très utilisée, mais a une limitation intrinsèque appelée le principe d’incertitude de Heisenberg-Gabor concernant la localisation en temps et en fréquence du signal. La méthode de réallocation a été proposée pour pallier cette limitation, mais elle est non inversible, c’est-à-dire qu’elle ne permet pas la reconstruction des modes à partir des transformées réallouées. Pour éliminer ce défaut, la méthode du synchoqueezing fondée sur la TFCT (FSST) a été développée. Néanmoins, cette méthode est seulement adaptée aux signaux contenant des modes de type chirp linéaires avec des amplitudes Gaussiennes modulées, ce qui constitue encore une limitation pour le traitement de signaux plus généraux. Pour mieux traiter de signaux comportant de très fortes modulations fréquentielles, la méthode du synchoqueezing d’ordre supérieur se fondant sur la TFCT (FSSTn) a été récemment proposé. Dans cet exposé, nous présenterons en détail des étapes principales du développement d’une telle technique et une application intéressante à l’analyse de signaux comportant de très fortes modulations fréquentielles : les ondes gravitationnelles.