University of Paris

About: University of Paris is a education organization based out in Paris, France. It is known for research contribution in the topics: Population & Medicine. The organization has 102426 authors who have published 174180 publications receiving 5041753 citations. The organization is also known as: Sorbonne.

Unsupervised Machine Translation Using Monolingual Corpora Only

Abstract: Machine translation has recently achieved impressive performance thanks to recent advances in deep learning and the availability of large-scale parallel corpora. There have been numerous attempts to extend these successes to low-resource language pairs, yet requiring tens of thousands of parallel sentences. In this work, we take this research direction to the extreme and investigate whether it is possible to learn to translate even without any parallel data. We propose a model that takes sentences from monolingual corpora in two different languages and maps them into the same latent space. By learning to reconstruct in both languages from this shared feature space, the model effectively learns to translate without using any labeled data. We demonstrate our model on two widely used datasets and two language pairs, reporting BLEU scores of 32.8 and 15.1 on the Multi30k and WMT English-French datasets, without using even a single parallel sentence at training time.

GALAXY STELLAR MASS ASSEMBLY BETWEEN 0.2 < z < 2 FROM THE S-COSMOS SURVEY*

Abstract: We follow the galaxy stellar mass assembly by morphological and spectral type in the COSMOS 2 deg^2 field. We derive the stellar mass functions and stellar mass densities from z = 2 to z = 0.2 using 196,000 galaxies selected at F_(3.6 μm) > 1 μJy with accurate photometric redshifts (σ_[(zphot−zspec)]/(1+zspec) = 0.008 at i^+ < 22.5). Using a spectral classification, we find that z ~ 1 is an epoch of transition in the stellar mass assembly of quiescent galaxies. Their stellar mass density increases by 1.1 dex between z = 1.5-2 and z = 0.8-1 (Δt ~ 2.5 Gyr), but only by 0.3 dex between z = 0.8-1 and z ~ 0.1 (Δt ~ 6 Gyr). Then, we add the morphological information and find that 80%-90% of the massive quiescent galaxies (logM ~ 11) have an elliptical morphology at z < 0.8. Therefore, a dominant mechanism links the shutdown of star formation and the acquisition of an elliptical morphology in massive galaxies. Still, a significant fraction of quiescent galaxies present a Spi/Irr morphology at low mass (40%-60% at logM ~ 9.5), but this fraction is smaller than predicted by semi-analytical models using a "halo quenching" recipe. We also analyze the evolution of star-forming galaxies and split them into "intermediate activity" and "high activity" galaxies. We find that the most massive "high activity" galaxies end their high star formation rate phase first. Finally, the space density of massive star-forming galaxies becomes lower than the space density of massive elliptical galaxies at z < 1. As a consequence, the rate of "wet mergers" involved in the formation of the most massive ellipticals must decline very rapidly at z < 1, which could explain the observed slow down in the assembly of these quiescent and massive sources.

Analysis of galaxy spectral energy distributions from far-UV to far-IR with CIGALE: studying a SINGS test sample

Abstract: Aims. Photometric data of galaxies covering the rest-frame wavelength range from far-UV to far-IR make it possible to derive galaxy properties with a high reliability by fitting the attenuated stellar emission and the related dust emission at the same time.Methods. For this purpose we wrote the code CIGALE (Code Investigating GALaxy Emission) that uses model spectra composed of the Maraston (or PEGASE) stellar population models, synthetic attenuation functions based on a modified Calzetti law, spectral line templates, the Dale & Helou dust emission models, and optional spectral templates of obscured AGN. Depending on the input redshifts, filter fluxes were computed for the model set and compared to the galaxy photometry by carrying out a Bayesian-like analysis. CIGALE was tested by analysing 39 nearby galaxies selected from SINGS. The reliability of the different model parameters was evaluated by studying the resulting expectation values and their standard deviations in relation to the input model grid. Moreover, the influence of the filter set and the quality of photometric data on the code results was estimated.Results. For up to 17 filters with effective wavelengths between 0.15 and 160 μ m, we find robust results for the mass, star formation rate, effective age of the stellar population at 4000 A, bolometric luminosity, luminosity absorbed by dust, and attenuation in the far-UV. Details of the star formation history (excepting the burst fraction) and the shape of the attenuation curve are difficult to investigate with the available broad-band UV and optical photometric data. A study of the mutual relations between the reliable properties confirms the dependence of star formation activity on morphology in the local Universe and indicates a significant drop in this activity at about 1011 towards higher total stellar masses. The dustiest galaxies in the SINGS sample are present in the same mass range. On the other hand, the far-UV attenuation of our sample galaxies does not appear to show a significant dependence on star formation activity.Conclusions. The results for our SINGS test sample demonstrate that CIGALE can be a valuable tool for studying basic properties of galaxies in the near and distant Universe if UV-to-IR data are available.

The CoRoT satellite in flight : description and performance

Abstract: Context. CoRoT is a space telescope dedicated to stellar seismology and the search for extrasolar planets. The mission is led by the CNES in association with French laboratories and has a large international participation. The European Space Agency (ESA), Austria, Belgium, and Germany contribute to the payload, and Spain and Brazil contribute to the ground segment. Development of the spacecraft, which is based on a PROTEUS low earth orbit (LEO) recurrent platform, commenced in October 2000, and the satellite was launched on December 27, 2006. Aims. The instrument and platform characteristics prior to launch have been described in ESA publication (SP-1306). In the present paper we explain the behaviour in flight, based on raw and corrected data. Methods. Five runs have been completed since January 2007. The data used here are essentially those acquired during the commissioning phase and from a long run that lasted 146 days. These enable us to give a complete overview of the instrument and platform behaviour for all environmental conditions. The ground based data processing is not described in detail because the most important method has been published elsewhere. Results. We show that the performance specifications are easily satisfied when the environmental conditions are favourable. Most of the perturbations, hence data corrections, are related to LEO perturbations: high energy particles inside the South Atlantic Anomaly (SAA), eclipses and temperature variations, and line of sight fluctuations due to the attitude control system. Straylight due to the reflected light from the earth, which is controlled by the telescope and baffle design, appears to be negligible.