LERMA UMR8112

Nous présenterons deux approches s’appuyant sur les réseaux de neurones profonds. La première approche utilise des méthodes de détection d’objets afin d'identifier de nouvelles nébuleuses planétaires à partir du catalogue IPHAS (The INT Photometric H_alpha Survey of the Northern Galactic Plane). La deuxième approche s’appuie quand à elle sur les travaux récents autour des auto-encodeurs variationnels qui utilisent un réseau de neurones pour l’apprentissage non supervisé de caractéristiques discriminantes. Cette approche est utilisée dans le cadre de l’estimation des formes 3D d’astéroïdes à partir des images radars en provenance des observatoires de Goldstone et Arecibo.


-

Accretion and ejection of matter is a process that plays a central role in the physics of Young Stellar Objects as well as of super-massive Black Holes in Active Galactic Nuclei. For both cases semi-analytical solutions can be obtained via a non-linear separation of the variables, which are the spherical distance and the magnetic flux function, in the relevant physical quantities. Using a self-similar solution as initial conditions, we performed 2.5D magneto-hydrodynamical simulations of the accretion-ejection system for Classical T Tauri Stars, YSO of about one solar mass. The collimated stellar jet and the accreting magnetosphere attain a quasi steady state in only few stellar rotations, confirming the robustness and stability of the self-similar solutions. Depending upon the accretion velocity, the density in the accretion column and the static magnetosphere, coronal episodic mass ejection can result from the interaction between the jet and the accretion column.

Spine jet modelling for relativistic outflows aroud Kerr Black Holes has been performed by applying meridional self-similarity technics to GRMHD equations for a perfect conducting ideal gas. We are able to describe the outflow near the super-massive central BH, and in particular to study the effects of the BH rotation. The model, characterized by 8 parameters, is based on a first order expansion of the governing general relativistic equations in the magnetic flux function around the symmetry axis of the system. I will present four enthalpy driven solutions with different field geometries and Lorentz factors, wherein the contribution of the Poynting flux is rather small. The jet power of the ultra-relativistic outflow solutions are of the same order as the one determined from numerical simulations conducted by several groups. Furthermore, our model is able to describe an incoming flow entering the BH horizon since at the stagnation radius, pairs are created from neutrinos or highly energetic photons coming from the disk. Coupling inflow and outflow models allows us to describe the MHD flow from the horizon of the BH up to infinity. We can estimate the different contributions of each of those processes: the energetic component at the BH horizon coming from the Blandford-Znajek effect or the generalized Penrose mechanism , and the energetic input due to the creation of pairs.

The new generation of wide-bandwidth high-resolution receivers turns
almost any radio observation into a spectral survey. In the case of
wide-field imaging of the interstellar medium, such a wealth of data
provides new diagnostic tools, but also poses new challenges in terms of
data processing and analysis.

The ORION-B project aims at observing 5 square degrees of the Orion B
molecular cloud, or about half of the cloud's surface, over the entire
3mm band. The emission of tens of molecular tracers have been mapped,
including CO isotopologues, HCO+, CN, HNC, N2H+, methanol, SO, CN...
Machine learning techniques have been applied to these maps, in order to
segment the molecular cloud into typical regions based on their
molecular emission, and to idenfify the most meaningful correlations of
different molecular tracers with each other and with physical quantities
such as density or dust temperature.

The spatial coverage, together with the spatial and spectral resolution,
also allow to characterize statistically the kinematics and dynamics of
the gas. The amount of momentum in the compressive and solenoidal
(rotational) modes of turbulence are retrieved, showing that the cloud
is dominated by solenoidal motions, with the compressive modes being
concentrated in two star-forming regions - which is in line with the
overall very low star formation efficiency of the cloud, and highlights
the role of compressive forcing in the star formation process. The
filamentary network of the molecular cloud also proves to have
particluarly low densities, and is very stable against gravitational
collapse and fragmentation, which also points at a young evolutionary
stage of the filaments.

Star-forming regions present rich infrared and millimeter spectra emitted by the gas exposed to the feedback of young stars. This emission is increasingly used to study the star formation cycle in other galaxies, but results from a complex interplay of physical and chemical processes : chemistry in the gas and on grain surfaces, (de)excitation processes of the atoms and molecules, heating and cooling balance,... Its understanding thus requires detailed astrochemical models that include the couplings between these processes. In this talk, I will present several examples where new modeling approaches of specific processes and their couplings proved crucial to solve persistent observational riddles : from the driving role of UV irradiation in the dynamics of photodissociation regions (PDR) to the efficient reformation of molecular hydrogen in these regions.