Archive Ouverte INSA Lyon
Not a member yet
75807 research outputs found
Sort by
A characteristic mapping method with source terms: Applications to ideal magnetohydrodynamics
No full textInternational audienceThis work introduces a generalized characteristic mapping method designed to handle non-linear advection with source terms. The semi-Lagrangian approach advances the flow map, incorporating the source term via the Duhamel integral. We derive a recursive formula for the time decomposition of the map and the source term integral, enhancing computational efficiency. Benchmark computations are presented for a test case with an exact solution and for two-dimensional ideal incompressible magnetohydrodynamics (MHD). Results demonstrate third-order accuracy in both space and time. The submap decomposition method achieves exceptionally high resolution, as illustrated by zooming into fine-scale current sheets. An error estimate is performed and suggests third order convergence in space and time, which is in agreement with the numerical results
Discrétisation de milieux multicouches à forts contrastes : qu'en est-il des interfaces ?
No full textWave propagation in multilayered media with high material contrasts poses significant numerical challenges, as large variations in wavenumbers lead to strong reflections and complex transmission of the incoming wave field. To address these difficulties, we employ a boundary integral formulation thereby avoiding volumetric discretization. In this framework, the accuracy of the numerical solution depends strongly on how the material interfaces are discretized. In this work, we demonstrate that standard meshing strategies based on resolving the maximum wavenumber across the domain become computationally inefficient in multilayered configurations, where high wavenumbers are confined to localized subdomains. Through a systematic study of multilayer transmission problems, we show that no simple discretization rule based on the maximum wavenumber or material contrasts emerges. Instead, the wavenumber of the background (exterior) medium plays a dominant role in determining the optimal boundary resolution. Building on these insights, we propose an adaptive approach that achieves uniform accuracy and efficient computation across multiple layers. Numerical experiments for a range of multilayer configurations demonstrate the scalability and robustness of the proposed approach
Multi-Score Reinforcement Learning for High-Tg Polyimide Design
No full textInternational audienceThis study explores strategies to guide the generation of polyimides with high glass transition temperatures (Tg > 750 K) through reinforcement learning. We present a systematic computational framework for analyzing and combining multiple scoring functions into a single score in reinforcement learning (RL) for molecular design. Rather than relying solely on a single scoring function based on a predictive model, we examine a range of complementary scores, including a novel naı̈ve high-Tg score and various Tanimoto similarity-based scores. We analyze these scores both individually and in combination with the predictive model-based score in order to assess their influence on the structural diversity and quality of the generated polymers. In addition, we investigate several methods for combining scores, such as arithmetic, geometric, and harmonic means, as well as a novel exponential–logarithmic function, referred to as ExpAgg. We evaluate how these aggregation strategies affect the outcomes of molecular generation across different reinforcement learning configurations. Our findings show that the choice of score combination method significantly impacts both the quality and diversity of generated polymers. The proposed ExpAgg achieves superior performance in multiple settings, revealing nontrivial interactions between score compatibility and model convergence. While the predictive model exhibits underestimation in the out-of-distribution region (>800 K), our multiscore framework successfully generates chemically reasonable high-Tg candidates. Based on these insights, we provide practical guidelines for selecting aggregation functions when fusing two scores. This case study on high-Tg polyimide generation demonstrates how score aggregation strategies influence molecular RL outcomes; broader generalizability to other molecular design tasks remains to be investigated. This work emphasizes the importance of moving beyond simple weighted averages in order to enhance targeted molecular design
Patch-based Representation and Learning for Efficient Deformation Modeling
No full textInternational audienc
La transition écologique dans les organisations de santé : retour d’expérience d’une formation-action à l’EHESP
No full textInternational audienc
Encyclopedia2GeoKG : un outil pour l'extraction d'informations et la génération de graphes de connaissances géo-historiques à partir d'articles encyclopédiques
No full textNational audienceLes dictionnaires et encyclopédies anciennes, comme celle de Diderot et d'Alembert (1751-1772), renferment des connaissances, notamment géographiques, qui sont précieuses pour étudier leur évolution au cours des derniers siècles. Les milliers d'articles dans ces oeuvres nécessitent cependant des outils automatisés pour extraire des informations fiables et structurées. Notre outil s'appuie sur des modèles entraînés spécifiquement pour les textes anciens et permet de construire un graphe RDF représentant les entités géographiques et leurs relations spatiales. Un prototype interactif de l'outil ainsi que les modèles sont disponibles sur HuggingFace, le code est disponible sur gitla
Augmentation and bulk edge correspondence for one dimensional aperiodic tight binding operators
No full textInternational audienceWe consider a particular class of 1D aperiodic models with the aim to understand how their internal degrees of freedom contribute to their topological invariants and the possible relations (correspondences) among them. In order to handle models with finite local complexity we introduce the principle of augmentation. This allows us to relate the values of the Integrated Density of States at gap energies for the bulk system to spectral flows. We consider two different augmentations. The first is based on the mapping torus construction. It leads to an alternative proof of the result that the gap labelling group of Bellissard coincides with that of Johnson-Moser. It furthermore allows for an interpretation of the spectral flow via boundary forces. The second augmentation applies to models obtained by the cut and project method where we find for 2-cut models two different spectral flows, one attached to the edge modes and related to the phason motion whereas the other is an augmented bulk invariant. Our approach is based on the well-established C * -algebraic approach to solid state physics and the description of topological invariants by K-theory and cyclic cocycles. We also present numerical simulations to illustrate our theorems
The Model's Language Matters: A Comparative Privacy Analysis of LLMs
No full textInternational audienceLarge Language Models (LLMs) are increasingly deployed in multilingual settings that process sensitive data, yet their scale and linguistic variability can amplify privacy risks. While prior privacy evaluations focus predominantly on English, we investigate how language structure shapes privacy leakage in LLMs trained on English, Spanish, French, and Italian medical corpora. We quantify six corpus-level linguistic indicators and evaluate vulnerability under three attack families: extraction, counterfactual memorization, and membership inference. Across languages, we find that leakage systematically tracks structural properties: Italian exhibits the strongest exposure, consistent with its highest redundancy and longer lexical units, whereas English shows the clearest membership separability, aligning with its higher syntactic entropy and stronger surface-identifiable cues. In contrast, French and Spanish remain comparatively more resilient overall, aided by higher morphological complexity. These results provide quantitative evidence that language matters for privacy leakage, motivating language-aware and structure-adaptive privacy-preserving mechanisms for multilingual LLM deployments
Joint reconstruction and pansharpening for high-resolution hyperspectral single-pixel imaging
No full textInternational audienceWe address the problem of single-pixel hyperspectral imaging, which requires balancing acquisition speed and spatial resolution. To improve spatial resolution when acquiring a small number of measurements in low-light conditions, we leverage side information from a high-resolution grayscale camera. Our joint reconstruction and fusion approach combines hyperspectral measurements and the grayscale image by minimizing a hand-crafted cost function that incorporates smooth spatial regularization and a low-rank approximation. Experiments on synthetic data show that our method improves spatial fidelity and per-pixel accuracy in photon-limited settings while preserving spectral alignment
