INRIA a CCSD electronic archive server
Not a member yet
122212 research outputs found
Sort by
Goal-oriented error estimation based on equilibrated flux reconstruction for the approximation of the harmonic formulations in eddy current problems
No full textInternational audienceIn this work, we propose an a posteriori goal-oriented error estimator for the harmonic A/Phi formulation arising in the modeling of eddy current problems, approximated by nonconforming finite element methods. It is based on the resolution of an adjoint problem associated with the initial one. For each of these two problems, a guaranteed equilibrated estimator is developed using some flux reconstructions. These fluxes also allow to obtain a goal-oriented error estimator that is fully computable and can be split in a principal part and a remainder one. Our theoretical results are illustrated by numerical experiments
A new tool for the simulation of controllers for retarded delay systems
No full textInternational audienceIn this paper we illustrate functionalities of a new software tool developed for reliableimplementation of all stabilizing controllers for retarded time delay systems. This work is thecontinuation of our paper in TDS2024, Bonnet et al. (2024); it demonstrates a Matlab/Simulinkrealization and how users interface with it. The paper also discusses some technical issues in theimplementation of the infinite dimensional parts of the controller blocks
On a general robust stability test based on the homological perturbation lemma
No full textInternational audienceWithin the lattice approach to synthesis problems, we show how a general unstructured robust stability test can be obtained directly by applying the homological perturbation lemma, a standard method developed in algebraic topology and homological algebra. This robust stability test generalizes and unifies various results from the robust control literature
Mixed precision implicit numerical schemes for systems of ordinary differential equations
No full textInternational audienceOrdinary differential equations (ODEs) are widely used to model complex systems in biology, which result from the interactions of a large number of cells or organisms. This can lead to a substantial system size. These complex interactions can quickly alter their behavior, and some biological systems are stiff when represented as ODEs [59]. Therefore, these stiff ODEs could benefit from implicit numerical schemes. However, each iteration of these schemes involves solving a large nonlinear system, typically using the Newton method. To guarantee the global convergence of this method, we use line search (LS) and trust region (TR) algorithms. In this paper, we introduce a new approach to accelerate the computation of the implicit schemes by using mixed precision arithmetic, combining float and double precision, within the LS and TR algorithms, as well as in the Newton method. This approach aims at balancing the performance of lower precision arithmetic with the accuracy of higher precision arithmetic. We give theoretical results that show the efficiency of our approach. Numerical experiments show that our approach, running in either sequential or in parallel with MPI, is up to twice as fast as the double precision approach with the same level of accuracy. These experiments also show that increasing the size of the ODEs does not impact the quality of our mixed precision solution
Legally validated evaluation framework for voice anonymization
No full textInternational audienceClassical speaker verification metrics used to evaluate voice anonymization systems, such as the equal error rate (EER), fail to properly quantify the residual re-identification risk. This paper introduces a new evaluation framework based on two metrics, Linkability and Singling Out, derived from the legal definitions in the Article 29 Working Party's Opinion 05/2014 on Anonymization Techniques endorsed by the European Data Protection Board (EDPB). Our framework translates these legal concepts into quantitative metrics for speech data. The proposed framework has been legally validated by the French Data Protection Authority. Experiments across various attack scenarios reveal that, while the EER remains stable, Linkability and Singling Out vary much more. This demonstrates that the residual privacy risk after anonymization is far more variable than indicated by the EER, underscoring the need for evaluation metrics that align with legal criteria
volesti: A C++ library for sampling and volume computation on convex bodies
No full textInternational audienc
Quantile Regression, Variational Autoencoders, and Diffusion Models for Uncertainty Quantification: A Spatial Analysis of Sub-seasonal Wind Speed Prediction
No full textInternational audienceThis study aims to improve the spatial representation of uncertainties when regressing surface wind speeds from large-scale atmospheric predictors for sub-seasonal forecasting. Sub-seasonal forecasting often relies on large-scale atmospheric predictors such as 500 hPa geopotential height (Z500), which exhibit higher predictability than surface variables and can be downscaled to obtain more localised information. Previous work by Tian et al. (2024) demonstrated that stochastic perturbations based on model residuals can improve ensemble dispersion representation in statistical downscaling frameworks, but this method fails to represent spatial correlations and physical consistency adequately. More sophisticated approaches are needed to capture the complex relationships between large-scale predictors and local-scale predictands while maintaining physical consistency. Probabilistic deep learning models offer promising solutions for capturing complex spatial dependencies. This study evaluates three probabilistic methods with distinct uncertainty quantification mechanisms: Quantile Regression Neural Network that directly models distribution quantiles, Variational Autoencoders that leverage latent space sampling, and Diffusion Models that utilise iterative denoising. These models are trained on ERA5 reanalysis data and applied to ECMWF sub-seasonal hindcasts to regress probabilistic wind speed ensembles. Our results show that probabilistic downscaling approaches provide more realistic spatial uncertainty representations compared to simpler stochastic methods, with each probabilistic model offering different strengths in terms of ensemble dispersion, deterministic skill, and physical consistency. These findings establish probabilistic downscaling as an effective enhancement to operational sub-seasonal wind forecasts for renewable energy planning and risk assessment
Dichromatic number of chordal graphs
No full textInternational audienceThe dichromatic number of a digraph is the minimum integer such that admits a -dicolouring, \textit{i.e.} a partition of its vertices into acyclic subdigraphs. We say that a digraph is a super-orientation of an undirected graph if is the underlying graph of . If does not contain any pair of symmetric arcs, we just say that is an orientation of .In this work, we give both lower and upper bounds on the dichromatic number of super-orientations of chordal graphs. In general, the dichromatic number of such digraphs is bounded above by the clique number of the underlying graph (because chordal graphs are perfect). However, this bound can be improved when we restrict the symmetric part of such a digraph.Let be a super-orientation of a chordal graph . Let be the undirected graph with vertex set in which is an edge if and only if both and belongs to . An easy greedy procedure shows . We show that this bound is best possible by constructing, for every fixed with , a super-orientation of a chordal graph such that , and .When (\textit{i.e.} is an orientation of ), we give another construction showing that this is tight even for orientations of interval graphs.Next, we show that with the maximum average degree of .Finally, we show that if contains no as a subgraph, then . We justify that this is almost best possible by constructing, for every fixed , a super-orientation of a chordal graph with clique number such that is a disjoint union of paths and .We also exhibit a family of orientations of cographs for which the dichromatic number is equal to the clique number of the underlying graph
Software Engineering Agents for Embodied Controller Generation : A Study in Minigrid Environments
No full textInternational audienceSoftware Engineering Agents (SWE-Agents) have proven effective for traditional software engineering tasks with accessible codebases, but their performance for embodied tasks requiring well-designed information discovery remains unexplored. We present the first extended evaluation of SWE-Agents on controller generation for embodied tasks, adapting Mini-SWE-Agent (MSWEA) to solve 20 diverse embodied tasks from the Minigrid environment. Our experiments compare agent performance across different information access conditions: with and without environment source code access, and with varying capabilities for interactive exploration. We quantify how different information access levels affect SWE-Agent performance for embodied tasks and analyze the relative importance of static code analysis versus dynamic exploration for task solving. This work establishes controller generation for embodied tasks as a crucial evaluation domain for SWE-Agents and provides baseline results for future research in efficient reasoning systems.Les agents d'ingénierie logicielle (SWE-Agents) ont prouvé leur efficacité pour les tâches traditionnelles d'ingénierie logicielle avec des bases de code accessibles, mais leurs performances pour les tâches incarnées nécessitant une découverte d'informations bien conçue restent inexplorées. Nous présentons la première évaluation approfondie des SWE-Agents sur la génération de contrôleurs pour les tâches incarnées, en adaptant Mini-SWE-Agent (MSWEA) pour résoudre 20 tâches incarnées diverses à partir de l'environnement Minigrid. Nos expériences comparent les performances des agents dans différentes conditions d'accès à l'information : avec et sans accès au code source de l'environnement, et avec des capacités variables d'exploration interactive. Nous quantifions l'impact des différents niveaux d'accès à l'information sur les performances des agents SWE pour les tâches incarnées et analysons l'importance relative de l'analyse statique du code par rapport à l'exploration dynamique pour la résolution des tâches. Ce travail établit la génération de contrôleurs pour les tâches incarnées comme un domaine d'évaluation crucial pour les agents SWE et fournit des résultats de référence pour les recherches futures sur les systèmes de raisonnement efficaces
Mappes difféomorphiques pour la réduction de modèle appliqué en aérodynamique
No full textParametric fluid dynamics simulations present significant computational challenges in aerodynamics, where high-fidelity computational fluid dynamics (CFD) models involve millions of degrees of freedom. Reduced-order models (ROMs) address this computational burden by constructing low-dimensional approximations of the full-order system. However, standard projection-based approaches exhibit poor performance for advectiondominated flows. A proposed solution is to rely on coordinate transformation mappings to enhance ROMs compression. This thesis addresses the issue of developing a general framework for diffeomorphic mappings to align aerodynamic structures for application in coordinate transformation-based ROMs. To answer this problem, we first establish the mathematical foundations of the method. For this, the mapping of interest is defined as the minimizer of an objective function. This constitutes the registration problem. The objective function combines a data misfit term based on point-set alignment of coherent structures with a regularization term derived from differential operators. The mapping is defined from a parametrization of a velocity field. This guarantees diffeomorphic transformations in bounded CFD domains under tangent boundary conditions. The minimization problem is solved using a finiteelement discretization, gradient-based optimization, and an Expectation-Maximization algorithm for automatic coherent-structure labeling. The methodology is validated on three representative test cases of increasing complexity: (i) coalescing Gaussian mixtures, to illustrate the alignment of merging structures; (ii) transonic Euler flow around a NACA0012 airfoil, demonstrating multi-structure registration; and (iii) viscous RANS flow around the ONERA M6 wing, showcasing the alignment of complex 3D lambda shocks. Across these cases, the proposed approach enhances ROM accuracy and demonstrates robustness to aerodynamic problems.Les simulations paramétriques de dynamique des fluides présentent des défis computationnels significatifs en aérodynamique. Les modèles de dynamique des fluides computationnelle (CFD) de haute fidélité impliquent des millions de degrés de liberté. Les modèles d’ordre réduit (ROM) résolvent cette charge computationnelle en construisant des approximations de faible dimension du système d’ordre complet. Cependant, les approches standard basées sur la projection linéaire des équations présentent de mauvaises performances pour les écoulements dominés par la convection. Une solution proposée consiste à s’appuyer sur des changements de coordonnées pour améliorer la compression des ROMs. Cette thèse aborde la question du calcul des transformations difféomorphes dans le cadre général afin d’aligner les structures aérodynamiques pour l’application de ROMs plus robustes. Pour répondre à ce problème, nous établissons d’abord les fondements mathématiques de la méthode. Pour cela, la transformation d’intérêt est définie comme le minimiseur d’une fonction objective. Ceci constitue le problème de recalage (ou d’alignement). La fonction objective combine un terme d’inadéquation des données, basé sur l’alignement d’ensembles de points de structures cohérentes, avec un terme de régularisation dérivé d’un opérateur différentiel. La transformation est définie à partir d’une paramétrisation d’un champ de vitesse. Ceci garantit des transformations difféomorphes dans les domaines CFD bornés grâce à des conditions aux limites tangentielles. Le problème de minimisation est résolu en utilisant une discrétisation par éléments finis, une optimisation basée sur le gradient et un algorithme d’Espérance-Maximisation pour l’étiquetage automatique des structures cohérentes. La méthodologie est validée sur trois cas de test représentatifs de complexité croissante : (i) des mélanges gaussiens coalescents, pour illustrer l’alignement de structures fusionnantes ; (ii) un écoulement Euler transsonique autour d’un profil NACA0012, démontrant le recalage multi-structures ; et (iii) un écoulement RANS visqueux autour de l’aile ONERA M6, présentant l’alignement de chocs en lambda dans des géométries 3D. À travers ces cas, l’approche proposée améliore la précision des ROM et démontre une robustesse aux problèmes aérodynamiques