Portail HAL-PSL
Not a member yet
293219 research outputs found
Sort by
Les chiens et les chats transportent les vers plats envahissants de jardin à jardin
No full textInternational audienc
A CNN encoder for modal phase reconstruction in Adaptive Optics systems
No full textPyramid wavefront sensing (pWFS) offers some of the best sensitivity for adaptive optics, but suffers from strong non-linearity. Modulation can extend linear range at the expense of sensitivity. Operating the pyramid without modulation is therefore attractive, but remains challenging. We develop a non-linear CNN reconstructor for pyramid wavefront sensing that maps pWFS images directly to a modal phase representation, and we assess simple hybrid methods combining it with a standard linear least-squares (LS) reconstructor. Using the end-to-end COMPASS simulator, we generate a large open-loop dataset spanning wide ranges of RMS and power spectra to train a compact CNN encoder. We then compare the CNN against an LS baseline and evaluate hybrid schemes in closed-loop simulations over a grid of guide-star magnitudes and Fried parameters, reporting long-exposure Strehl at lambda=1.6 um with controller gain re-optimized per method and bin. We also report preliminary offline bench tests on SCExAO. The CNN reduces open-loop reconstruction error and, in closed loop, outperforms LS across most (magnitude, r_0) conditions, with the largest gains for faint stars where it often closes the loop while LS does not. In very strong turbulence, LS can exceed the CNN; in these cases the hybrid methods are necessary to surpass LS, with a second-stage NN performing best. Inference is real-time capable and the hybrid overhead is negligible. Bench snapshots on SCExAO show successful correction of small static/slow perturbations, with instability for stronger/faster cases. A compact CNN can be trained to perform modal reconstruction for a non-modulated pWFS and improves performance over a classical linear reconstructor in most regimes. When the CNN alone is not optimal, simple hybrid methods achieve the best performance, suggesting a practical way to exploit the pyramid’s sensitivity without modulation
Problèmes de la liberté d’expression en France et en Grèce
No full textInternational audienceLa Grèce a fait face ces dernières années à une crise économique d’une violence inouïe, à la montée d’une extrème droite ouvertement pronazie, tandis que la religion tient encore une place essentielle dans la vie quotidienne. Toutes ces caractéristiques confrontent la liberté d’expression à des problèmes tantôt spécifiques, tantôt semblables à ceux que connaît la France, mais auxquels des réponses différentes sont apportées. Certains chapitres de ce volume opèrent une véritable comparaison entre les droits français et hellénique sur des points particuliers. Plusieurs auteurs grecs présentent et analysent par ailleurs des débats méconnus en France, tandis que les contributions françaises jettent un regard renouvelé et approfondi sur des questions voisines. Peut-on détourner le drapeau national, se moquer d’une divinité ou d’un président, provoquer à la violence, prier dans la rue ou promouvoir la gestation pour autrui ? Certains des problèmes principaux posés par la liberté d’expression reçoivent ici une analyse inédite
The Well-Tempered Classifier: Some Elementary Properties of Temperature Scaling
No full textTemperature scaling is a simple method that allows to control the uncertainty of probabilistic models. It is mostly used in two contexts: improving the calibration of classifiers and tuning the stochasticity of large language models (LLMs). In both cases, temperature scaling is the most popular method for the job. Despite its popularity, a rigorous theoretical analysis of the properties of temperature scaling has remained elusive. We investigate here some of these properties. For classification, we show that increasing the temperature increases the uncertainty in the model in a very general sense (and in particular increases its entropy). However, for LLMs, we challenge the common claim that increasing temperature increases diversity. Furthermore, we introduce two new characterisations of temperature scaling. The first one is geometric: the tempered model is shown to be the information projection of the original model onto the set of models with a given entropy. The second characterisation clarifies the role of temperature scaling as a submodel of more general linear scalers such as matrix scaling and Dirichlet calibration: we show that temperature scaling is the only linear scaler that does not change the hard predictions of the model
Multi-Objective Categorical Deep Q-Networks
No full textInternational audienceMotivated by recent advances in distributional reinforcement learning on the one hand and Multi-Objective Reinforcement Learning (MORL) on the other, we propose MO-CDQN, a value based algorithms that, given a possibly non-linear scalarization function, learn the policy with maximal expected scalarized return. Leveraging the Kantorovich-Rubenstein duality we prove the theoretical validity of our method for Lipschitz-continuous scalarization function. We establish that the state-action return distributions learned by our algorithm converge to a fixed point whose expected scalarized return is optimal. Our approach is then extended to propose a first valuebased multi-policy algorithm for solving MORL problems under the expected scalarized return criterion. The proposed algorithms are tested on several environments from the MO-gymnasium benchmark. The results are promising and show that, on the one hand, our algorithm learns policies better than those obtained by existing approaches in the literature while requiring fewer interactions with the environments. On the other hand, given a set of scalarization functions, our multi-policy takes advantage of its off-policy nature to successfully optimize several policies concurrently and efficiently provide a set of policies each one optimal for a given scalarization function.</div
Brownian dynamics simulations of electric double-layer capacitors with tunable metallicity
No full textInternational audienceWe introduce an efficient description of electrodes, characterized by their Thomas–Fermi screening length inside the metal, for Brownian dynamics (BD) simulations of capacitors. Within a Born–Oppenheimer approximation for the electron charge density inside the electrodes, we derive the effective many-body potential for ions in an implicit solvent between Thomas–Fermi electrodes, taking into account the constraints of applied voltage and of global electro-neutrality of the system, as well as the 2D periodic boundary conditions along the electrode surfaces. We derive the average charge and the fluctuation–dissipation relation for the differential capacitance, highlighting the contribution of the fluctuations of the net ionic dipole moment, as well as those from the solvent polarization and of the electron density, whose fluctuations are suppressed within the Born–Oppenheimer description. We demonstrate the relevance of this model by validating its predictions against known results for the force on ions as a function of the ion-surface distance in simple geometries. The equilibrium ionic density profiles from BD simulations are in excellent agreement with those from an explicit electrode model for perfect metals and are obtained at a significantly lower computational cost. Finally, we discuss with the present model the effect of the Thomas–Fermi screening length on the equilibrium ionic density profiles and the capacitance. While limited to parallel plate capacitors, the present simulation method allows us to consider larger systems, lower concentrations, and longer time scales than molecular simulations in order to predict the electrochemical properties of Thomas–Fermi capacitors and correlate them with the ion dynamics
Efficacy, safety, pharmacokinetics, immunogenicity, and serum neutralizing activity of AZD7442 (tixagevimab-cilgavimab) in patients hospitalized with COVID-19: long-term results from the DisCoVeRy trial
No full textInternational audienceObjectives: To report long-term clinical efficacy, safety, pharmacokinetics, immunogenicity and seroneutralization results of AZD7442 (monoclonal antibodies tixagevimab-cilgavimab) in patients hospitalized with COVID-19.Methods: In this phase 3, double-blind, randomized, multicentre trial, hospitalized adults with PCR-confirmed SARS-CoV-2 infection were randomly assigned 1:1 to receive AZD7442 or placebo, and followed-up until day 456, with repeated blood sample collections until day 365. Clinical endpoints included clinical status, mortality, rehospitalization, SARS-CoV-2 reinfection, and adverse events. Antidrug antibodies and serum drug concentrations were measured. Analyses were performed on the modified intention-to-treat (mITT) populations, defined as participants who actually received the intervention.Results: Between April 28, 2021, and June 23, 2022, 237 participants were randomly assigned to AZD7442 (n = 127) or placebo (n = 110), and 123 participants actually received AZD7442. Participants were infected with pre-Omicron variants in 58.8% (133/226) of cases, versus 33.2% (75/226) of Omicron BA1, BA2, or BA5, and 8% (18/226) missing data. There was no significant difference in the distribution of the 7-point ordinal scale between the AZD7442 and placebo groups, either on day 15 (primary endpoint) (OR = 0.93 [0.54-1.61], p 0.81), or any other time point. Significantly more rehospitalizations occurred between discharge and day 456 among participants who received AZD7442 in the global mITT population (OR = 2.04 [1.03-4.05], p 0.04), but not in the antigen-positive mITT population (OR = 1.78 [0.80-3.94], p 0.15). No significant differences were observed in mortality, SARS-CoV-2 reinfection, or adverse events. In the AZD7442 group, 12 of 87 participants (13.8%) had treatment-emergent antidrug antibodies versus 5 of 69 (7.2%) in the placebo group (OR = 2.02 [0.66-6.14], p 0.21). Serum drug concentrations were detectable up to day 365 for all sampled participants (35/35). Neutralizing antibody titres were significantly higher in the AZD7442 group up to day 180.Conclusions: AZD7442 did not demonstrate any clinical benefit and was safe up to 15 months. This study also provides valuable data on the pharmacokinetics, immunogenicity, and neutralizing activity of AZD7442 in patients hospitalized with COVID-19
Evidence for the collective nature of radial flow in Pb+Pb collisions with the ATLAS detector
No full textInternational audienceAnisotropic flow and radial flow are two key probes of the expansion dynamics and properties of the quark-gluon plasma (QGP). While anisotropic flow has been extensively studied, radial flow, which governs the system's radial expansion, has received less attention. Notably, experimental evidence for the global and collective nature of radial flow has been lacking. This Letter presents the first measurement of transverse momentum () dependence of radial flow fluctuations () over GeV, using a two-particle correlation method in Pb+Pb collisions at TeV. The data reveal three key features supporting the collective nature of radial flow: long-range correlation in pseudorapidity, factorization in , and centrality-independent shape in . The comparison with a hydrodynamic model demonstrates the sensitivity of to bulk viscosity, a crucial transport property of the QGP. These findings establish a new, powerful tool for probing collective dynamics and properties of the QGP
The Schouten-Nijenhuis bracket, codifferential of products and generalized interior products of p-forms
No full textInternational audienceIdentities pertaining to the de Rham codifferential δ in differential geometry are scattered in the literature. This article gathers such formulas involving usual differential operators (Lie derivative, Schouten-Nijenhuis bracket, etc.), while adding some new ones using a natural extension of the interior product, to provide a compact handy summary