HAL-Polytechnique
Not a member yet
51406 research outputs found
Sort by
On the hardness of Reinforcement Learning with Transition Look-Ahead
No full textInternational audienceWe study reinforcement learning (RL) with transition look-ahead, where the agent may observe which states would be visited upon playing any sequence of actions before deciding its course of action. While such predictive information can drastically improve the achievable performance, we show that using this information optimally comes at a potentially prohibitive computational cost. Specifically, we prove that optimal planning with one-step look-ahead () can be solved in polynomial time through a novel linear programming formulation. In contrast, for , the problem becomes NP-hard. Our results delineate a precise boundary between tractable and intractable cases for the problem of planning with transition look-ahead in reinforcement learning
Uncovering identifiability of epidemiological models: basic reproduction number and complementary data streams
No full textMathematical models of infectious disease dynamics are routinely fitted to surveillance data to estimate epidemiological parameters and inform public health decisions. Such data are typically discrete and noisy, but before attempting estimation, it is essential to ask whether the model structure itself permits unique parameter identification at least under perfect (continuous, noise-free) observations. This mathematical property of a model with respect to observation(s), known as structural identifiability, serves as a necessary precondition for reliable inference, since a model that fails this test cannot yield unique parameter estimates even from perfect data. In this study, we systematically investigate structural identifiability in various classes of compartmental epidemic models and establish two main findings. First, we present and deploy a methodology for assessing structural identifiability of epidemiological quantities of interest and demonstrate that the basic reproduction number exhibits identifiability across diverse model structures- including models with multiple transmission pathways and host-vector dynamics- even when individual parameters are not uniquely identifiable. These findings challenge the assumption that complete model identifiability is necessary for reliable epidemiological inference and suggest reformulating the central question from "is the model identifiable?" to "are the quantities that matter for the decision-making identifiable?" Second, we prove that incorporating minimal complementary data, as little as a single time-point measurement from an additional state variable, can make otherwise nonidentifiable models globally identifiable. This result has direct implications for surveillance design: rather than putting limited resources into frequent monitoring of multiple data streams or relying on external parameter estimates that may be uncertain or contex-dependent, public health systems can strategically prioritize collecting high-quality complementary measurements
H.E.S.S. detection and multi-wavelength study of the 1 blazar PKS 034627
No full textInternational audiencePKS 0346-27 is a Low Synchrotron Peaked (LSP) blazar at redshift 0.991. The very-high-energy (VHE, E > 100 GeV) spectra of blazars are always affected by absorption by the Extragalactic Background Light (EBL) and subsequently, no blazars have been detected in VHE -rays at redshifts exceeding 1. Extending the redshift range of VHE-detected blazars to will yield insights into the cosmological evolution of both the VHE blazar population and the EBL. This is the goal of a target-of-opportunity (ToO) programme by H.E.S.S. to observe flaring high-redshift () blazars. We report on H.E.S.S. ToO and multi-wavelength observations of the blazar PKS 034627. Along with H.E.S.S., simultaneous data from {\it Fermi}-LAT, {\it Swift} (XRT and UVOT), and ATOM have been analysed and modelled using single-zone leptonic and hadronic models. PKS~0346-27 has been detected by H.E.S.S at a significance of 6.3 during one night, on 3 November 2021, while for other nights before and after this day, upper limits on the VHE flux are determined. No evidence for intra-night -ray variability has been found. A flare in high-energy (HE, ~MeV) -rays detected by {\it Fermi}-LAT preceded the H.E.S.S. detection by 2 days. A fit with a single-zone emission model to the contemporaneous spectral energy distribution during the detection night was possible with a proton-synchrotron-dominated hadronic model, requiring a proton-kinetic-energy-dominated jet power temporarily exceeding the source's Eddington limit, although alternative (e.g. multi-zone) models can not be ruled out. A one-zone leptonic model is, in principle, also able to fit the flare-state SED, however, requiring implausible parameter choices, in particular, extreme Doppler and bulk Lorentz factors of
Illustrator's Depth: Monocular Layer Index Prediction for Image Decomposition
No full textInternational audienceWe introduce Illustrator’s Depth, a novel definition of depth that addresses a key challenge in digital content creation: decomposing flat images into editable, ordered layers. Inspired by an artist’s compositional process, illustrator’s depth infers a layer index for each pixel, forming an interpretable image decomposition through a discrete, globally consistent ordering of elements optimized for editability. We also propose and train a neural network using a curated dataset of layered vector graphics to predict layering directly from raster inputs. Our layer index inference unlocks a range of powerful downstream applications. In particular, it significantly outperforms state-of-the-art baselines for image vectorization while also enabling high-fidelity text-to-vector-graphics generation, automatic 3D relief generation from 2D images, and intuitive depth-aware editing. By reframing depth from a physical quantity to a creative abstraction, illustrator’s depth prediction offers a new foundation for editable image decomposition
Physique de la matière condensée: [résumé des cours et travaux : 2021-2022]
No full textInternational audienc
Russia’s Invasion of Ukraine and Perceived Intergenerational Mobility in Czechia and Uruguay. An Unexpected Event During Survey Design Study
No full textCan large-scale geopolitical shocks causally reshape how individuals perceive their intergenerational mobility? We address this question by examining the immediate impact of Russia's invasion of Ukraine on mobility perceptions using a quasi-experimental design. Leveraging World Values Survey data collected in Czechia and Uruguay during the invasion, we implement a regression discontinuity design that compares respondents interviewed immediately before and after February 24, 2022. This unexpected event during survey design approach addresses key limitations of correlational studies by exploiting quasi-random variation in exposure to the shock. Across multiple specifications -- including bandwidth sensitivity analyses, logit models, demographic controls, and robust bias-corrected estimates -- we find consistent null effects. The invasion produced no discernible immediate impact on either upward or downward mobility perceptions in either country, despite Czechia's geographic proximity and historical ties to the region. These findings complement prior correlational evidence showing associations between war concerns and mobility beliefs one year after the invasion, suggesting that such effects may unfold gradually rather than instantaneously, or that correlational patterns reflect selection into worry rather than immediate causal impacts. Our study demonstrates the value of combining multiple methods and measures when studying how geopolitical crises shape social perceptions
Language Barriers and Healthcare Uptake. Causal Evidence From Immigrants in France
No full textImmigrants often arrive with better health than natives of their host country, a phenomenon known as the ‘healthy immigrant effect.’ This advantage diminishes over time, one reason being unequal healthcare access. Our study examines whether communication barriers contribute to these inequalities in a healthcare system that rarely accommodates an increasingly diverse and multilingual population. We expect that limited French proficiency restricts equitable access to care. Using the Trajectoires et Origines 2 survey conducted in 2019 in France among over 27,000 individuals---oversampling immigrant-origin populations---we investigate how language shapes healthcare use. France provides a relevant case, as immigrants come from diverse origins, including former colonies where French is an official language, generating substantial variation in proficiency. Findings show significant language-related disparities in medical visits, except for generalist practitioners. To address possible endogeneity bias in the relationship between language and healthcare uptake, we construct an instrument combining age at arrival in France and linguistic distance between French and the mother tongue. This method reveals a significant language effect only for visits to psychologists or psychiatrists. Theoretically, we propose a quantitative strategy to assess language barriers’ effects on immigrants’ healthcare access. Practically, we highlight the need for linguistically inclusive health services
Simultaneous estimation of radiance and its sensitivities to radiative properties in a spherical-heterogeneous atmospheric radiative transfer model by Monte Carlo method: Application to Titan
No full textInternational audienceWe propose a control variates technique to reduce the variance of null-collision Monte Carlo algorithms used for solving the Radiative Transfer Equation (RTE) in highly heterogeneous media. The method complements the classical spatially partitioned overestimate approach by additionally recording the minimum absorption coefficient within each voxel during preprocessing. During path tracing, the attenuation due to this minimum absorption is evaluated analytically, while the residual part is handled by path-samplings. This analytical treatment significantly improves convergence particularly in strongly absorbing media such as the planetary atmospheres in infrared absorbing band. The mathematical equivalence between the original and control-variates estimators is demonstrated, and numerical applications for Earth's and Titan's atmospheres confirm the expected variance reduction.</div
Search for nucleon decay via and in 0.484 Mton-year of Super-Kamiokande data
No full textInternational audienceWe present the results of searches for nucleon decays via and using a 0.484 Mtyr exposure of Super-Kamiokande I-V data covering the entire pure water phase of the experiment. Various improvements on the previous 2014 nucleon decay search, which used an exposure of 0.173 Mtyr, are incorporated. The physics models related to pion production and nuclear interaction are refined with external data, and a more comprehensive set of systematic uncertainties, now including those associated with the atmospheric neutrino flux and pion production channels is considered. Also, the fiducial volume has been expanded by 21%. No significant indication of a nucleon decay signal is found beyond the expected background. Lower bounds on the nucleon partial lifetimes are determined to be ~yr for and ~yr for at 90% confidence level
Measurements of ttW differential cross sections and the leptonic charge asymmetry at = 13 TeV
No full textInternational audienceMeasurements of properties of top quark-antiquark pair production in association with a W boson in proton-proton collisions at a center-of-mass energy of 13 TeV are presented, using a data sample corresponding to an integrated luminosity of 138 fb, recorded by the CMS experiment at the CERN LHC. Events are selected based on the presence of either two leptons with the same electric charge or three leptons, and multiple jets and b-tagged jets. We present measurements of differential production cross sections as a function of kinematic variables sensitive to different aspects of the process modeling, using a multivariate discriminator in the two-lepton selection region and a simple selection-based method in the three-lepton region. The normalized cross section measurements are generally consistent with the standard model expectations, while we observe larger values compared to the expectations in the absolute cross section measurements, consistent with previous inclusive cross section measurements. In addition, we measure the leptonic charge asymmetry of this process, obtaining an observed value of = 0.19, consistent with the expectation of 0.0850.006 predicted by next-to-leading order simulations