Portail HAL Ensta
Not a member yet
11080 research outputs found
Sort by
IA générative, société et éducation: En quoi l'IA générative représente-elle un enjeu dans la formation des citoyens ?
No full textAlors qu'ils sont apparus très récemment, les grands systèmes d'IA générative ont déjà aujourd'hui un impact majeur sur la société, dans les domaines culturels, politiques, économiques, environnementaux et éducatifs. Leurs usages se développent en particulier très vite et massivement chez les jeunes, que ce soit dans le contexte scolaire ou en dehors. La vitesse de ce développement est telle que les études scientifiques permettant de mieux comprendre les usages et leurs impacts sont encore très rares car elles nécessitent un temps incompressible de mise en place et de vérification. On peut dire que globalement, d'un point de vue scientifique, c'est une terra incognita: on sait peu et nombreuses sont les questions ouvertes. Néanmoins, ce qui est sûr, c'est que la formation des futurs citoyens aux mécanismes et aux enjeux cognitifs, sociaux et culturels de l'IA générative est un enjeu majeur: Comment ces systèmes fonctionnent-ils ? Comment bien les utiliser ? Quels sont les défis (e.g. désinformation, uniformisation de la pensée, emplois, coûts environnementaux) et les opportunités (e.g. aide aux apprentissages et à la création, favoriser l'accès et la diffusion de cultures diverses et pour des personnes diverses, amélioration de la productivité) l'accès pour la société
Complete suppression of N lasing by the nonadiabatic molecular alignment effect in femtosecond filaments
No full textInternational audienceThe nonadiabatic molecular alignment effect is commonly found helpful in enhancing nonlinear processes such as high-order harmonics and strong-field ionization. In this work, an opposite phenomenon is observed, in which the nonadiabatic molecular alignment effect prepared with a linearpolarized prepulse strongly suppresses the generation of N lasing induced by circularly-polarized 800 nm femtosecond laser pulse in filament plasma. The presence of a weak prepulse periodically suppresses the lasing at each revival timing of the rotational wave packet of the N molecule. The underlying mechanism of the lasing signal suppression is attributed to sudden change of the polarization ellipticity of the pump laser pulse. Theoretical simulations by numerically solving the time-dependent Schrödinger equation with the molecular alignment effect included confirm sensitive changes of the polarization ellipticity of the pump laser at every alignment revival, which largely supports our interpretation
Contributions to Efficient Finite Element Solvers for Time-Harmonic Wave Propagation Problems
No full textThe numerical simulation of wave propagation phenomena is of paramount importance in many scientific and engineering disciplines. Many time-harmonic problems can be solved with finite elements in theory, but the computational cost is a strong constraint that limits the size of the problems and the accuracy of the solutions in practice. Ideally, solution techniques should provide the best accuracy at minimal computational cost for real-world problems. They should take advantage of the power of modern parallel computers, and they should be as easy as possible to use for the end user. In this HDR thesis, contributions are presented on three topics: the improvement of domain truncation techniques (i.e. high-order absorbing boundary conditions and perfectly matched layers), the acceleration of substructuring and preconditioning techniques based on domain decomposition methods (i.e. non-overlapping domain decomposition methods with interface conditions based on domain truncation techniques), and the design of a new hybridization approach for efficient discontinuous finite element solvers
Integration of artificial intelligence technologies in the defense sector and the effect of National Innovation System performance on the level of integration
No full textInternational audienceThis paper employs graph theory to assess the extent of integration of artificial intelligence (AI) technologies within defense activities and investigates how the performance of the national innovation system (NIS) influences this integration. The analysis utilizes data from 33 countries with defense industries, observed from 1990 to 2020. Empirical findings indicate that the United States (U.S.) leads globally, with a significant gap between the U.S. and other countries. NIS performance increases the level of integration of AI technologies in defense activities, suggesting that policies aimed at strengthening NIS performance should have positive externalities on defense activities in terms of integrating AI technologies. Technological diversification, knowledge localization, and originality are key dimensions of NIS performance that significantly enhance the integration of AI technologies within defense activities. They exhibit similar average marginal effects, suggesting comparable impacts. The cycle time of technologies has an inverted-U shaped relationship with the level of integration
Reconciling Spatial and Temporal Abstractions for Goal Representation
No full textInternational audienceGoal representation affects the performance of Hierarchical Reinforcement Learning (HRL) algorithms by decomposing the complex learning problem into easier subtasks. Recent studies show that representations that preserve temporally abstract environment dynamics are successful in solving difficult problems and provide theoretical guarantees for optimality. These methods however cannot scale to tasks where environment dynamics increase in complexity i.e. the temporally abstract transition relations depend on larger number of variables. On the other hand, other efforts have tried to use spatial abstraction to mitigate the previous issues. Their limitations include scalability to high dimensional environments and dependency on prior knowledge.In this paper, we propose a novel three-layer HRL algorithm that introduces, at different levels of the hierarchy, both a spatial and a temporal goal abstraction. We provide a theoretical study of the regret bounds of the learned policies. We evaluate the approach on complex continuous control tasks, demonstrating the effectiveness of spatial and temporal abstractions learned by this approach
Deep sea cold seeps are a sink for mercury and source for methylmercury
No full textInternational audienceThe effect of seafloor cold seeps on the biogeochemical cycling of mercury (Hg) remains enigmatic. Here we demonstrate substantial enrichments of mercury and methylmercury, along with the presence of microbes capable of metabolizing mercury in sediments of the Haima cold seep, South China Sea, by analyzing mercury and methylmercury concentrations, mercury isotopic composition analyses and metagenomic analyses of sediment cores. Compared to the reference area, the sediments in the upper sediment column of the active-seep area were 2.4 times enriched in Hg and 10.5 times in methylmercury. The slope of the capital delta ratio of mercury 199 to mercury 201 (Δ 199 Hg/Δ 201 Hg) with 1.23 ± 0.10 in the active-seep area indicate the occurrence of dark redox reactions. Genes related to mercury methylation ( hgcA ), demethylation ( merB ) and reduction ( merA ) were phylogenetically associated with several bacterial and archaeal linages. We roughly estimated an additional 2,835 Mg mercury and 9 Mg methylmercury are stored in cold seep globally. In summary, we propose that cold seeps globally function as a previously unrecognized sink for mercury and source for methylmercury in the deep ocean
Capsize criteria in beam seas: Melnikov analysis vs. safe basin erosion
No full textInternational audienceWe study the problem of capsizing of a rolling ship in harmonic and random beam seas, by means of the Melnikov method and safe basin calculations. In the random excitation case, we consider an integro-differential Cummins-type model equation that takes into account the hydrodynamic memory. The non-linear restoring moment is modeled by a high-order polynomial and the Melnikov criteria are evaluated numerically. We also examine a variant of the classical Melnikov method in which the damping terms are incorporated into the unperturbed system and the perturbation takes a modified form. We compare the theoretical predictions with direct (Monte-Carlo) simulations of the safe basins for two existing ships. In the random excitation case, we quantify the erosion by means of a mean integrity index. For weak damping, the classical and modified Melnikov curves coincide and are in good agreement with the onset of the safe basin erosion. As damping increases the Melnikov curves become less and less conservative with respect to the onset of erosion. Finally, the potential of the Melnikov method to provide a ship classification tool is discussed
Phase-field simulation and coupled criterion link echelon cracks to internal length in antiplane shear
No full textInternational audienceThis paper provides a comprehensive numerical analysis of daughter crack localization in pure antiplane shear. Although antiplane shear fracture is important in various industrial applications, understanding the morphology of the resulting fragmentation remains challenging. The paper develops innovative phase-field models to induce the facets using a small spatial variation in the toughness field and examines the impact of numerical and material parameters on the newly formed daughter cracks’ shape and spacing. Through meticulous comparison to the coupled criterion, the paper reveals a compelling connection between the internal length-scale of damage regularization, Irwin’s length and the facet crack spacing. Furthermore, the effect of Poisson’s ratio on the crack form and spacing is investigated: the results reveal a significant influence and showcase comparable initiation distances between the numerical simulations and experimental measurements in pure antiplane loading
Ustensiles et espaces culinaires de la Protohistoire au début du XXe siècle, pré-actes du colloque international Corpus, Dijon, 4-7 juin 2024
No full textInternational audienc
Clustering data for the Optimal Classication Tree Problem
No full textInternational audienceSolving the optimal classification tree problem enables to compute classifiers which are both interpretable and efficient. Most of the exact methods for this problem are based on on a Mixed Integer Linear Program (MILP) formulation. However, the efficiency of MILP solvers generally does not allow these formulations to be solved directly, once the dataset exceeds a critical size. To address this challenge, we propose in this paper an iterative exact algorithm than handles medium-sized datasets from the state-of-the-art. The basic idea is to start by solving a MILP formulation on a small subset of data points representative of the considered dataset. Then, the subset is iteratively extended until global optimality of the initial problem is reached.A key feature is to compute relevant initial subsets of data points. For this, we introduce the concept of data-partitions and design several algorithms to compute them. We then define two MILP formulations to compute optimal classification trees on data-partitions. We prove that combining our iterative algorithm with our first formulation enables to obtain an optimal solution of the original problem. We also propose an alternative method based on the second formulation which is significantly faster.We present extensive computational experiments to compare our algorithms with state-of-the-art approaches. We show that our methods constitute the best compromise between in-sample accuracy and interpretability