University of Grenoble

About: University of Grenoble is a education organization based out in Saint-Martin-d'Hères, France. It is known for research contribution in the topics: Population & Large Hadron Collider. The organization has 25658 authors who have published 45143 publications receiving 909760 citations.

Multi-modal Transformer for Video Retrieval

Abstract: The task of retrieving video content relevant to natural language queries plays a critical role in effectively handling internet-scale datasets. Most of the existing methods for this caption-to-video retrieval problem do not fully exploit cross-modal cues present in video. Furthermore, they aggregate per-frame visual features with limited or no temporal information. In this paper, we present a multi-modal transformer to jointly encode the different modalities in video, which allows each of them to attend to the others. The transformer architecture is also leveraged to encode and model the temporal information. On the natural language side, we investigate the best practices to jointly optimize the language embedding together with the multi-modal transformer. This novel framework allows us to establish state-of-the-art results for video retrieval on three datasets. More details are available at http://thoth.inrialpes.fr/research/MMT.

Centrality determination of Pb-Pb collisions at root s(NN)=2.76 TeV with ALICE

Abstract: This publication describes the methods used to measure the centrality of inelastic Pb-Pb collisions at a center-of-mass energy of 2.76 TeV per colliding nucleon pair with ALICE. The centrality is a key parameter in the study of the properties of QCD matter at extreme temperature and energy density, because it is directly related to the initial overlap region of the colliding nuclei. Geometrical properties of the collision, such as the number of participating nucleons and the number of binary nucleon-nucleon collisions, are deduced from a Glauber model with a sharp impact parameter selection and shown to be consistent with those extracted from the data. The centrality determination provides a tool to compare ALICE measurements with those of other experiments and with theoretical calculations.

A circRNA from SEPALLATA3 regulates splicing of its cognate mRNA through R-loop formation

Abstract: Circular RNAs (circRNAs) are a diverse and abundant class of hyper-stable, non-canonical RNAs that arise through a form of alternative splicing (AS) called back-splicing. These single-stranded, covalently-closed circRNA molecules have been identified in all eukaryotic kingdoms of life1, yet their functions have remained elusive. Here, we report that circRNAs can be used as bona fide biomarkers of functional, exon-skipped AS variants in Arabidopsis, including in the homeotic MADS-box transcription factor family. Furthermore, we demonstrate that circRNAs derived from exon 6 of the SEPALLATA3 (SEP3) gene increase abundance of the cognate exon-skipped AS variant (SEP3.3 which lacks exon 6), in turn driving floral homeotic phenotypes. Toward demonstrating the underlying mechanism, we show that the SEP3 exon 6 circRNA can bind strongly to its cognate DNA locus, forming an RNA:DNA hybrid, or R-loop, whereas the linear RNA equivalent bound significantly more weakly to DNA. R-loop formation results in transcriptional pausing, which has been shown to coincide with splicing factor recruitment and AS2-4. This report presents a novel mechanistic insight for how at least a subset of circRNAs probably contribute to increased splicing efficiency of their cognate exon-skipped messenger RNA and provides the first evidence of an organismal-level phenotype mediated by circRNA manipulation.

Planck 2015 results: XIX. Constraints on primordial magnetic fields

Abstract: We predict and investigate four types of imprint of a stochastic background of primordial magnetic fields (PMFs) on the cosmic microwave background (CMB) anisotropies: the impact of PMFs on the CMB temperature and polarization spectra, related to their contribution to cosmological perturbations; the effect on CMB polarization induced by Faraday rotation; magnetically-induced non-Gaussianities and related non-zero bispectra; and the magnetically-induced breaking of statistical isotropy. We present constraints on the amplitude of PMFs derived from different combinations of Planck data products, depending on the specific effect that is analysed. Overall, Planck data constrain the amplitude of PMFs to less than a few nanogauss, with different bounds depending on the considered model. In particular, individual limits coming from the analysis of the CMB angular power spectra, using the Planck likelihood, are B1Mpc < 4:4 nG (where B1Mpc is the comoving field amplitude at a scale of 1 Mpc) at 95 % confidence level, assuming zero helicity, and B1Mpc < 5:6 nG when we consider a maximally helical field. For nearly scaleinvariant PMFs we obtain B1Mpc < 2:1 nG and B1Mpc < 0:7 nG if the impact of PMFs on the ionization history of the Universe is included in the analysis. From the analysis of magnetically-induced non-Gaussianity we obtain three different values, corresponding to three applied methods, all below 5 nG. The constraint from the magnetically-induced passive-tensor bispectrum is B1Mpc < 2:8 nG. A search for preferred directions in the magnetically-induced passive bispectrum yields B1Mpc < 4:5 nG, whereas the the compensated-scalar bispectrum gives B1Mpc < 3 nG. The analysis of the Faraday rotation of CMB polarization by PMFs uses the Planck power spectra in EE and BB at 70 GHz and gives B1Mpc < 1380 nG. In our final analysis, we consider the harmonic-space correlations produced by Alfv´ en waves, finding no significant evidence for the presence of these waves. Together, these results comprise a comprehensive set of constraints on possible PMFs with Planck data.