HAL-CEA
Not a member yet
145450 research outputs found
Sort by
Sparse factor analysis for categorical data with the group-sparse generalized singular value decomposition
No full textInternational audienceCorrespondence analysis, multiple correspondence analysis, and their discriminant counterparts (i.e., discriminant simple correspondence analysis and discriminant multiple correspondence analysis) are methods of choice for analyzing multivariate categorical data. In these methods, variables are integrated into optimal components computed as linear combinations whose weights are obtained from a generalized singular value decomposition (GSVD) that integrates specific metric constraints on the rows and columns of the original data matrix. The weights of the linear combinations are, in turn, used to interpret the components, and this interpretation is facilitated when components are 1) pairwise orthogonal and 2) when the values of the weights are either large or small but not intermediate—a configuration called a simple or a sparse structure. To obtain such simple configurations, the optimization problem solved by the GSVD is extended to include new constraints that implement component orthogonality and sparse weights. Because multiple correspondence analysis represents qualitative variables by a set of binary columns in the data matrix, an additional group constraint is added to the optimization problem in order to sparsify the whole set of columns representing one qualitative variable. This method—called group-sparse GSVD (gsGSVD)—integrates these constraints in a new algorithm via an iterative projection scheme onto the intersection of subspaces where each subspace implements a specific constraint. This algorithm is described in details, and we show how it can be adapted to the sparsification of simple and multiple correspondence analysis (as well as their barycentric discriminant analysis versions). This algorithm is illustrated with the analysis of four different data sets—each illustrating the sparsification of a particular CA-based method
Normally-off 1200 V GaN-on-Si recess MOSc-HEMT thanks to AlGaN back barrier
No full textInternational audienceThis paper reports for the first time a novel AlGaN and Mg doped GaN Back Barrier (AlGaN BB and Mg-BB respectively) on fully recessed GaN MOS High Electron Mobility Transistors (MOS-HEMTs) fabricated on 200 mm Si substrates. These BB were introduced to increase and stabilize the transistor threshold voltage (V) under harsh test conditions. The impact of the Al percentage (Al%) in the AlGaN BB was evaluated, revealing a parasitic transistor that was suppressed with careful optimization of the AlGaN BB properties. Then, the optimized AlGaN BB was compared to the Mg-BB, showing both improved ON state resistance (RON) and V > +1 V. Dynamic characterizations were carried out as well and demonstrated superior VTH and RON stability thanks to the novel AlGaN BB. The transistors voltage rating was also studied showing a 1550 V buffer-limited maximum breakdown voltage, fulfilling the specifications for 1200 V applications
Microstructural features governing the effective thermal conductivity of Cu-25Cr sintered composites
No full textInternational audienceMedium voltage vacuum interrupters are high-performance current interruption devices in the contemporary energy transition and electrification. The current interruption performance is directly correlated to the properties of their electrical contacts made of Cu-Cr alloys. Therefore, enhancing their performance to respond to the continuously increasing demand for higher current, higher power, and high voltage applications inherently requires the optimization of the microstructure of the Cu-Cr alloys to increase their electrical and thermal conductivity. Herein, we unveil the microstructural features governing the effective thermal conductivity of Cu-25Cr sintered composites, a subject of much less scientific attention than their electrical conductivity due to the complex microstructure-thermal conduction relationships. We coupled advanced 3D characterization techniques, namely X-ray computed tomography and atom probe tomography, with experimental and full-field numerical investigation of the effective thermal conductivity for three Cu-25Cr sintered composites having different final relative density (94, 96, and 98%). We demonstrate the synergistic effect of solid solution, interfacial thermal resistance, phase distribution on the effective thermal conductivity. Interfacial pores hinder the thermal conduction across phases. The effective thermal conductivity also decreases due to elements in solid-solution that diffused in the Cu matrix during sintering, along with interfacial thermal resistance across Cu/Cr phases. Using a full-field numerical approach, we unravel a microstructure heterogeneity-induced heat flux anisotropy contributing to the anisotropy in effective thermal conductivity.</div
Effect of post thermal-quench shattered pellet injection on runaway electron generation in ITER
No full textInternational audienc
Wasserstein normalized autoencoder for anomaly detection
No full textInternational audienceA novel anomaly detection algorithm is presented. The Wasserstein normalized autoencoder (WNAE) is a normalized probabilistic model that minimizes the Wasserstein distance between the learned probability distribution -- a Boltzmann distribution where the energy is the reconstruction error of the autoencoder -- and the distribution of the training data. This algorithm has been developed and applied to the identification of semivisible jets -- conical sprays of visible standard model particles and invisible dark matter states -- with the CMS experiment at the CERN LHC. Trained on jets of particles from simulated standard model processes, the WNAE is shown to learn the probability distribution of the input data in a fully unsupervised fashion, such that it effectively identifies new physics jets as anomalies. The model consistently demonstrates stable, convergent training and achieves strong classification performance across a wide range of signals, improving upon standard normalized autoencoders, while remaining agnostic to the signal. The WNAE directly tackles the problem of outlier reconstruction, a common failure mode of autoencoders in anomaly detection tasks
On the impacts of halo model implementations in Sunyaev-Zeldovich cross-correlation analyses
No full textInternational audienceStatistical studies of the circumgalactic medium (CGM) using Sunyaev-Zeldovich (SZ) observations offer a promising method of studying the gas properties of galaxies and the astrophysics that govern their evolution. Forward modeling profiles from theory and simulations allows them to be refined directly off of data, but there are currently significant differences between the thermal SZ (tSZ) observations of the CGM and the predicted tSZ signal. While these discrepancies could be real, they could also be the result of decisions in the forward modeling used to build statistical measures from theory. In order to see effects of this, we compare an analysis utilizing halo occupancy distributions (HODs) implemented in halo models to simulate the galaxy distribution against previous studies, which weighted their results to match the CMASS galaxy sample, which contains nearly one million galaxies, mainly centrals of group-sized halos, selected for relatively uniform stellar mass across redshifts between 0.4 < z < 0.7. We review some of the implementation differences that can account for changes, such as miscentering, one-halo/two-halo cutoff radii, and mass ranges, all of which will need to be given the proper attention in future high-signal-to-noise studies. We find that our more thorough model predicts a signal with a 33% improved fit than the one from previous studies on the exact same sample. Additionally, we find that modifications that change the satellite fraction even by just a few percent, such as editing the halo mass range and certain HOD parameters, result in strong changes in the final signal. Although significant, this discrepancy from the modeling choices is not large enough to completely account for the existing disagreements between simulations and measurements
Euclid preparation. Spatially resolved stellar populations of local galaxies with Euclid: a proof of concept using synthetic images with the TNG50 simulation
No full textInternational audienceThe European Space Agency's Euclid mission will observe approximately 14,000 of the extragalactic sky and deliver high-quality imaging for many galaxies. The depth and high spatial resolution of the data will enable a detailed analysis of stellar population properties of local galaxies. In this study, we test our pipeline for spatially resolved SED fitting using synthetic images of Euclid, LSST, and GALEX generated from the TNG50 simulation. We apply our pipeline to 25 local simulated galaxies to recover their resolved stellar population properties. We produce 3 types of data cubes: GALEX + LSST + Euclid, LSST + Euclid, and Euclid-only. We perform the SED fitting tests with two SPS models in a Bayesian framework. Because the age, metallicity, and dust attenuation estimates are biased when applying only classical formulations of flat priors, we examine the effects of additional priors in the forms of mass-age- relations, constructed using a combination of empirical and simulated data. Stellar-mass surface densities can be recovered well using any of the 3 data cubes, regardless of the SPS model and prior variations. The new priors then significantly improve the measurements of mass-weighted age and compared to results obtained without priors, but they may play an excessive role compared to the data in determining the outcome when no UV data is available. The spatially resolved SED fitting method is powerful for mapping the stellar populations of galaxies with the current abundance of high-quality imaging data. Our study re-emphasizes the gain added by including multiwavelength data from ancillary surveys and the roles of priors in Bayesian SED fitting. With the Euclid data alone, we will be able to generate complete and deep stellar mass maps of galaxies in the local Universe, thus exploiting the telescope's wide field, NIR sensitivity, and high spatial resolution
Immunogenicity of single-chain antibodies: germlining of a VHH lowers T-cell activation from epitopes in FR2 and CDR regions
No full textInternational audienceSingle-chain antibodies (scAbs), derived from camelid antibodies, have gained attention as therapeutic candidates due to their small size and perceived low immunogenicity, but recent studies have reported immune responses to several scAbs. To better understand their immunogenicity, we investigated the T-cell responses induced by VHH76, a VHH-Fc engineered to target the SARS-CoV-2 RBD, along with its humanized and germlined variants. The humanized variant contains six human substitutions, while the germlined variant was obtained by screening of a combinatorial library of the VHH76 sequences, comprising human and wild-type substitutions at 12 different positions. The germlined variant finally contains 16 human substitutions. All VHH76 variants triggered CD4 T-cell responses from healthy donors, with the germlined VHH76 showing significantly reduced T-cell stimulation. Two epitope regions were identified: one overlapping CDR3 and another spreading from CDR1 to CDR2. Additional human substitutions at the VHH-conserved positions in FR2 compromised the biological properties of the germlined VHH76 and did not seem to reduce clearly the risk of T-cell response. In conclusion, using a sensitive T-cell assay, we showed that T cells specific for VHH76 variants were detected in the blood of healthy donors and that the frequency of responding T cells diminished with germlining. While epitopes in CDR3 are linked to VHH76 specificity, modifying the conserved FR2 region presents challenges for reducing VHH76 immunogenicity. This study contributes to the understanding of VHH76 immunogenicity and offers insights into strategies to mitigate immune responses
Developments and implementation of the spent nuclear fuel recycling process PUMAS at the ATALANTE facility
No full textInternational audienc
Roadmap on carbon molecular nanostructures in space
No full textInternational audienceIn this roadmap article, we consider the main challenges and recent breakthroughs in understanding the role of carbon molecular nanostructures in space and propose future avenues of research. The focus lies on small carbon-containing molecules up to fullerenes, extending to even larger, more complex organic species. The roadmap contains forty contributions from scientists with leading expertize in observational astronomy, laboratory astrophysics/chemistry, astrobiology, theoretical chemistry, synthetic chemistry, molecular reaction dynamics, material science, spectroscopy, graph theory, and data science. The concerted interdisciplinary combination of the state-of-the-art of these astronomical, laboratory, and theoretical studies opens up new ways to advance the fundamental understanding of the physics and chemistry of cosmic carbon molecular nanostructures and touches on their wider relevance and impact in nanotechnology and catalysis