HAL-CentraleSupelec
Not a member yet
77624 research outputs found
Sort by
From Ethical Discourse to Empirical Evidence: How Ethics Is Operationalized in Studies on Generative AI in Higher Education
No full textPaper (long) accepted in AIED 2026 27th International Conference on AI in Education, Seoul Republic of Korea.International audienceThe rapid adoption of generative artificial intelligence (GenAI) in higher education has raised widespread ethical concerns, particularly in computer science (CS) education. While issues such as academic integrity, fairness, bias, and responsibility are frequently discussed, it remains unclear how these concerns are empirically examined in existing research. This paper presents a systematic analysis of how ethical considerations are addressed in peer-reviewed empirical studies on GenAI use in higher CS education. Using a structured selection protocol, we analyze a corpus of empirical studies published since the public release of large language models, focusing on how ethical dimensions are defined, studied, and methodologically grounded. Our analysis reveals a substantial gap between ethical discourse and empirical practice: although ethics is often mentioned, only a small subset of studies integrates ethical concerns as a core analytical dimension with explicit empirical grounding. Most studies rely on indirect proxies or address ethics implicitly, without clear operational definitions or evaluative frameworks. Based on these findings, we propose an empirically grounded categorization of ethical dimensions as addressed in current research and outline methodological directions for more robust integration of ethics into future empirical studies on GenAI-supported CS education.</div
Reduced-Rank Mutual Coupling Representation and Experimental Estimation for Large RIS
No full textInternational audiencePhysics-consistent optimization of reconfigurable intelligent surfaces (RISs) is thwarted in practice by the difficulty of experimentally estimating the mutual coupling (MC) between RIS elements. For large RISs, experimental MC estimation is fundamentally challenging because of the quadratic scaling of the number of unknowns with the number of RIS elements. In this letter, we present a generic and flexible reduced-rank MC representation that allows wireless practitioners to choose a trade-off between model complexity and accuracy. We experimentally validate the direct reduced-rank MC estimation for a 100-element RIS in three radio environments (rich scattering, attenuated scattering, free space). We observe a strong environmental dependence of the influence of rank reduction on accuracy. Model-based performance evaluations highlight that the importance of MC awareness in optimization depends strongly on the radio environment
TWO-DIMENSIONAL INFRARED EMISSIVITY OF A SEA SURFACE PERTURBED BY A CIRCULAR CYLINDER
No full textInternational audienceThe sea surface intrinsic infrared emission is describedby its emissivity. This measure can be evaluated fromtwo ways (i) Analytically (AN), where the statistical average ofthe local emissivity is done analytically by assuming a Gaussiansurface slope distribution (ii) Or by a numerical process (NUM)of Monte Carlo, meaning that the average emissivity is computedby generating numerous statistical sea heights surfaces. Our aimis to obtain the emissivity of a 2D sea surface perturbed bya submarine’s mast (periscope) crossing the free surface. Thisperturbation of the calm water free surface by a circular cylinderis simulated by a fluid mechanics software and superimposed toa background sea state that follows the Elfouhaily et al. seaspectrum. Thus, by comparing this emissivity with the one of abackground sea, the periscope could be detected
Insights into the stability analysis of 2-by-2 linear weakly hyperbolic systems
No full textIn this paper, we address the exponential stability analysis of steady states for one-dimensional linear weakly hyperbolic systems of conservation laws and balance laws, characterized by non-diagonalizable coefficient matrices. We derive sufficient conditions to ensure exponential stability, conducting the analysis both in the Laplace domain for a system of conservation laws and in the time domain using Lyapunov-based techniques. Numerical examples are presented to demonstrate the effectiveness of the proposed methods
La Fabrique des Jumeaux Numériques: verrous scientifiques, technologiques et perspectives du programme EDT
No full textNational audienceThe current transition from traditional physical systems to software-driven ecosystems is profoundly transforming engineering practices and increasing the demand for accessible and scalable technologies. Digital twins (DTs) are emerging as a foundational concept, enabling the separation of high-level decision-making from physical operations while maintaining a close connection to the real system. By combining deductive models from engineering and inductive models derived from data, they provide powerful capabilities for analysis, simulation, prediction, and optimization of systems. Despite this potential, their implementations often remain fragmented, domain-specific, and costly to deploy and maintain. This article synthesizes the scientific and technological challenges associated with digital twin engineering and outlines the objectives of the Engineering Digital Twins (EDT) program, which aims to establish a rigorous and reliable approach to their development, deployment, and operation.La transition des systèmes physiques traditionnels vers des écosystèmes pilotés par logiciel transforme profondément les pratiques d'ingénierie. Les jumeaux numériques (JN) émergent comme un concept structurant permettant de dissocier la prise de décision de haut niveau des opérations physiques tout en maintenant un lien étroit avec le système réel. En combinant des modèles déductifs issus de l'ingénierie et des modèles inductifs dérivés des données, ils offrent des capacités puissantes d'analyse, de simulation, de prédiction et d'optimisation des systèmes. Malgré ce potentiel, leurs implémentations restent souvent fragmentées, spécifiques à un domaine et coûteuses à déployer et à maintenir. Cet article propose une synthèse des verrous scientifiques et technologiques associés à l'ingénierie des jumeaux numériques et présente les objectifs du programme Engineering Digital Twins (EDT), qui vise une approche rigoureuse et fiable de leur développement, de leur déploiement et de leur exploitation
Output feedback stabilization of an ODE-Kuramoto-Sivashinsky PDE cascade
No full textInternational audienceThis paper studies the problem of output feedback stabilization of an ODE-Kuramoto-Sivashinsky PDE cascade. The control input applies to the ODE while the ODE output enters into the Kuramoto-Sivashinsky PDE through the left Dirichlet trace. The measurement is solely performed on the PDE subsystem. The adopted approach relies on spectral reduction techniques through a detailed study of the eigenelements of the ODE-PDE cascade. This allows to derive explicit, necessary, and sufficient modal controllability and observability conditions for the system. Under these conditions, we report an exponentially stabilizing output feedback control strategy for a PDE state that is evaluated in either L 2 or H 2 norm. The obtained stabilization results are global in the case of a linear Kuramoto-Sivashinsky PDE and are local in the nonlinear setting
Path-conditioned training: a principled way to rescale ReLU neural networks
No full textDespite recent algorithmic advances, we still lack principled ways to leverage the well-documented rescaling symmetries in ReLU neural network parameters. While two properly rescaled weights implement the same function, the training dynamics can be dramatically different. To offer a fresh perspective on exploiting this phenomenon, we build on the recent path-lifting framework, which provides a compact factorization of ReLU networks. We introduce a geometrically motivated criterion to rescale neural network parameters which minimization leads to a conditioning strategy that aligns a kernel in the path-lifting space with a chosen reference. We derive an efficient algorithm to perform this alignment. In the context of random network initialization, we analyze how the architecture and the initialization scale jointly impact the output of the proposed method. Numerical experiments illustrate its potential to speed up training
Recent Advances on Starch-Based Biomaterials: A Review
No full textInternational audienceStarch's versatility inspires biomaterials for biomedical uses, customized through modifications, blending, or substituents. Diligent efforts have been dedicated to the development of starch-based biomaterials, leveraging the material's inherent biocompatibility and biodegradability, while aligning with environmentally sustainable considerations. While promising, most studies lack in vivo data and scalability assessments. In many cases, the reported advances are restricted to in vitro evaluations with limited information on long-term performance, clinical translation, and large-scale manufacturing feasibility in both economic and operational terms. This review furnishes an up-to-date synthesis of information available in the literature concerning recent breakthroughs in utilizing starch as a biomaterial, primarily focusing on advancements in areas such as wound dressings, drug delivery systems, the creation of scaffolds for regenerative medicine, and applications in tissue engineering. Advances have been made, with biomaterials presenting adequate biodegradability rates, active functions, good biocompatibility, and mechanical properties. However, it is noted that most research has not yet reached in vivo evaluations and lacks notions of large-scale production, in both economic and operational terms
Statistical Consistency of Discrete-to-Continuous Limits of Determinantal Point Processes
No full textWe investigate the limiting behavior of discrete determinantal point processes (DPPs) towards continuous DPPs when the size of the set to sample from goes to infinity. We propose a non-asymptotic characterization of this limit in terms of the concentration of statistics associated to these processes, which we refer to as “weak coherency”. This allows to translate statistical guarantees from the limiting process to the original, discrete one. Our main result describes sufficient conditions for weak coherency to hold. In particular, our study encompasses settings where both the kernel of the continuous process and its underlying space are inaccessible, or when the discrete marginal kernel is a noisy version of its continuous counterpart. We illustrate our theory on several examples. We prove that a discrete multivariate orthogonal polynomial ensemble can be used to produce coresets strictly smaller than independent sampling for the same error. We propose a process achieving repulsive sampling on an unknown manifold from a set of points sampled from an unknown density. Finally, we show that continuous DPPs can be obtained as limits on random graphs with Bernoulli edges, even when only observing the graph structure. We obtain interesting byproduct results along the way
Approximation theory for distant Bang calculus
No full textApproximation semantics capture the observable behaviour of λ-terms; Böhm Trees and Taylor Expansion are its two central paradigms, related by the Commutation Theorem. While well understood in Call-by-Name (CbN), these notions were only recently developed for Call-by-Value (CbV), motivating the search for a unified approximation framework. The Bang-calculus provides such a framework, subsuming both CbN and CbV through linear-logic translations while providing robust rewriting properties. We develop the approximation semantics of dBang-the Bang-calculus with explicit substitutions and distant reductions-by defining Böhm trees and Taylor expansion and establishing their fundamental properties. Via translations, our results recover the CbN and CbV cases within a single unifying framework capturing infinitary and resource-sensitive semantics